A privacy-preserving anti-fraud system and method based on machine learning large models and ASR

Abstract

This invention relates to the fields of communication services and cloud communication technology, specifically disclosing a privacy-preserving anti-fraud system based on machine learning big data models and Automatic Speech Recognition (ASR). The system includes a data statistics and preprocessing module, a machine learning module, an automatic speech recognition module, an automatic inspection module, a call content recognition and judgment module, a service alarm module, and a service management module. This invention also discloses a privacy-preserving anti-fraud method based on machine learning big data models and ASR, employing the aforementioned anti-fraud system. This invention's privacy-preserving anti-fraud system and method, based on machine learning big data models and ASR, can proactively mine characteristics of fraudulent call behavior, automatically intercept high-risk fraudulent numbers, and effectively curb fraudsters from using privacy-preserving number platforms to commit fraud.

Description

Technical Field

[0001] This invention relates to the fields of communication services and cloud communication technology, and in particular to a privacy number anti-fraud system and method based on machine learning large models and ASR. Background Technology

[0002] As platform clients increasingly prioritize user privacy, the volume of business for privacy number operation platforms has surpassed tens of millions of transactions daily. Facing such massive amounts of data presents significant challenges in risk control, anomaly detection, and call tracking. Fraudsters exploit this vulnerability, using sophisticated and difficult-to-detect scripts to defraud users on privacy number platforms. Current technology lacks effective grouping of abnormal call records, leaving call tracking incomplete.

[0003] Current anti-fraud measures primarily rely on issuing fraud tickets after the fact, manually reviewing suspicious calls, and then blacklisting the caller and recipient numbers and freezing customer accounts. These methods are inefficient and slow, failing to effectively intercept fraudulent calls in a timely manner and posing significant security vulnerabilities. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems mentioned above, and provides a privacy number anti-fraud system and method based on machine learning large model and ASR, which can proactively discover the characteristics of fraudulent call behavior, automatically block high-risk fraudulent numbers, and effectively curb fraudsters from using privacy number platforms to commit fraud.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a privacy-preserving anti-fraud system based on a large machine learning model and ASR, comprising:

[0006] The data statistics and preprocessing module is responsible for refining the basic call detail record data in the database, completing the entire data extraction, transformation, and loading process to achieve data preprocessing;

[0007] Machine learning module: responsible for performing cluster analysis on preprocessed data, predicting and classifying call groups based on key indicator data, and dividing calls into normal call groups and abnormal call groups;

[0008] The automatic speech recognition module automatically recognizes the abnormal call groups predicted by the machine learning module and outputs the text results.

[0009] The automatic inspection module has the function of creating scheduled tasks and can pull call recordings classified as fraudulent call groups from the machine learning module to the automatic speech recognition module. Then, the recognized text content is scheduled to the keyword recognition and big model judgment tasks, so that the two models can perform tasks periodically and learn from the text content.

[0010] The call content recognition and judgment module has keyword recognition and large model judgment functions. The keyword recognition function can identify the dialogue text content of abnormal call groups and judge whether there is high-risk fraud-related content in the dialogue. The large model judgment function can perform semantic understanding on the text results output by the automatic speech recognition module and judge whether it meets the characteristics of fraud-related calls.

[0011] The business alarm module uses the process engine to aggregate machine learning data and content recognition results, assess the risk level and classification, and then push the alarm content to the business control module.

[0012] The business management module responds with different management strategies based on the alarm level and alarm category of the business alarm module, while also supplementing and maintaining the customer profile system.

[0013] Preferably, the key indicator data includes call duration, connection rate, and short call percentage. Calls with a duration of less than 10 seconds, a connection rate of less than 40%, and a short call percentage of more than 50% are classified as abnormal call groups.

[0014] Preferably, the risk levels are divided into Level 1 risk, Level 2 risk, and Level 3 risk, where Level 1 risk is prohibited or violated regulations, Level 2 risk is suspected fraud, Level 3 risk is high-risk business, and Level 4 risk is malicious harassment.

[0015] Preferably, the control strategy includes: when the operation receives alarms at each level, the priority of manual review of business is arranged according to the risk level, that is, the business of level 1 risk and level 2 risk is reviewed first. If the review result is correct, the business of level 1 risk and level 2 risk customers is directly frozen and taken offline. If the review result of level 3 risk and level 4 risk is correct, the business is taken offline as appropriate after communicating with the sales in charge.

[0016] This application also discloses a privacy-preserving account anti-fraud method based on a large machine learning model and ASR, which employs the aforementioned anti-fraud system and includes the following steps:

[0017] (1) The basic call detail record information is extracted, transformed and loaded by the data statistics preprocessing module, and key call feature index data is output to the machine learning module.

[0018] (2) The call feature index data output by the data preprocessing module is clustered by the machine learning module to divide the call groups into normal call groups and abnormal call groups.

[0019] (3) The abnormal call group call recording files output by machine learning are recognized as text content through the automatic speech recognition module;

[0020] (4) The keyword recognition function of the call content recognition and judgment module is used to identify keywords in the text content. If the text content matches the keyword, the matched keyword is extracted. If the text content does not match the keyword, the classification is manually reviewed to see if it is correct. If the classification is incorrect, the model is optimized. If the classification is correct, new keywords are labeled. The large model judgment function of the call content recognition and judgment module is used to judge whether the text content matches the abnormal call. If the text content matches the abnormal call, the judgment result is output. If the text content does not match the abnormal call, the judgment is manually reviewed to see if it is correct. If the judgment is incorrect, the model is optimized. Then the text learning results and machine learning results are aggregated.

[0021] (5) The alarm level is judged by the business alarm module after the aggregation results of machine learning and content recognition, and the alarm level and category are output. The business control module makes corresponding policy control based on the alarm level and category, and at the same time supplements and maintains the customer profile system.

[0022] Preferably, the model optimization method involves manually reviewing inaccurate call cases to determine whether the specific call scenario of the call case is consistent with the actual reported scenario. If they are consistent, keywords in that scenario are manually selected and the new keywords are maintained in the corresponding scenario keyword library. If they are inconsistent, the keywords are maintained in the keyword library that does not match the reported scenario.

[0023] The beneficial effects are as follows: Compared with existing technologies, the privacy number anti-fraud system and method of the present invention, based on machine learning big data and ASR, utilizes machine learning based on indicator data and big data models to support automatic speech recognition technology. This dual approach results in high efficiency in screening abnormal call groups and more accurate identification and judgment of abnormal call content. Through preprocessing of tens of millions of call records, the machine learning module classifies call groups according to their call characteristics, quickly screening out suspected fraudulent call groups. Keyword recognition and big data model judgment perform secondary verification and judgment on the call content of these suspected fraudulent call groups. This enables more accurate identification of fraudulent call numbers and accelerates the control and response speed of abnormal call groups, thereby curbing the occurrence of fraud cases. Attached Figure Description

[0024] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, wherein:

[0025] Figure 1 This is a schematic diagram of the privacy number anti-fraud system architecture based on a large machine learning model and ASR according to the present invention;

[0026] Figure 2Flowchart for identifying and processing fraudulent call groups;

[0027] Figure 3 This is a timeline diagram for managing fraudulent calls. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a central component. When a component is described as "connected to" another component, it can be directly connected to the other component or may have a central component. When a component is described as "set on" another component, it can be directly set on the other component or may have a central component. When a component is described as "set in the middle," it is not simply set in the exact center, as long as it is not set within the area defined by both ends being in the middle. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0031] like Figures 1 to 3 As shown, this application discloses a privacy-preserving anti-fraud system for mobile phone numbers based on a large machine learning model and ASR, including:

[0032] The data statistics and preprocessing module is responsible for refining the basic call detail record data in the database, completing the entire data extraction, transformation, and loading process to achieve data preprocessing;

[0033] Machine Learning Module: Responsible for performing cluster analysis on preprocessed data, predicting and classifying call groups based on key indicator data, and dividing calls into normal call groups and abnormal call groups. Key indicator data includes call duration, connection rate, and proportion of short calls. Based on these key indicators, abnormal call groups are predicted. Calls with a duration of less than 10 seconds, a connection rate of less than 40%, and a proportion of short calls of more than 50% are considered abnormal call groups.

[0034] The automatic speech recognition module automatically recognizes the abnormal call groups predicted by the machine learning module and outputs the text results.

[0035] The automatic inspection module has the function of creating scheduled tasks and can pull call recordings classified as fraudulent call groups from the machine learning module to the automatic speech recognition module. Then, the recognized text content is scheduled to the keyword recognition and big model judgment tasks, so that the two models can perform tasks periodically and learn from the text content.

[0036] The call content recognition and judgment module has keyword recognition and large model judgment functions. The keyword recognition function can identify the dialogue text content of abnormal call groups and judge whether there is high-risk fraud-related content in the dialogue. The large model judgment function can perform semantic understanding on the text results output by the automatic speech recognition module and judge whether it meets the characteristics of fraud-related calls.

[0037] The business alarm module uses the process engine to aggregate machine learning data and content recognition results to assess risk level and classification, and then pushes the alarm content to the business control module. The risk level is divided into three levels: Level 1 risk, Level 2 risk, and Level 3 risk. Level 1 risk is prohibited or violated regulations, Level 2 risk is suspected fraud, Level 3 risk is high-risk business, and Level 4 risk is malicious harassment.

[0038] The business control module responds with different control strategies based on the alarm level and alarm category of the business alarm module, and also supplements and maintains the customer profile system. The alarm level categories are as follows:

[0039] Control strategy: When the operations team receives alarms at each level, they will arrange the priority of manual review of the business according to the risk level. That is, they will prioritize the review of Level 1 and Level 2 risk businesses. If the review results are correct, the business of Level 1 and Level 2 risk customers will be frozen and taken offline. If the review results of Level 3 and Level 4 risk businesses are correct, they will be taken offline as appropriate after communicating with the sales staff in charge.

[0040] Furthermore, this application also discloses a privacy-preserving anti-fraud method based on a large machine learning model and ASR, employing the aforementioned anti-fraud system and including the following steps:

[0041] (1) The basic call detail record information is extracted, transformed and loaded by the data statistics preprocessing module, and key call feature index data is output to the machine learning module.

[0042] (2) The call feature index data output by the data preprocessing module is clustered by the machine learning module to divide the call groups into normal call groups and abnormal call groups.

[0043] (3) The abnormal call group call recording files output by machine learning are recognized as text content through the automatic speech recognition module;

[0044] (4) The keyword recognition function of the call content recognition and judgment module is used to identify keywords in the text content. If the text content matches the keyword, the matched keyword is extracted. If the text content does not match the keyword, the classification is manually reviewed to see if it is correct. If the classification is incorrect, the model is optimized. If the classification is correct, new keywords are labeled. The large model judgment function of the call content recognition and judgment module is used to judge whether the text content matches the abnormal call. If the text content matches the abnormal call, the judgment result is output. If the text content does not match the abnormal call, the judgment is manually reviewed to see if it is correct. If the judgment is incorrect, the model is optimized. Then the text learning results and machine learning results are aggregated. The optimization method is to manually review inaccurate call cases and judge whether the specific call scenario of the call case is consistent with the actual reported scenario. If they are consistent, the keywords in the scenario are manually selected and the new keywords are maintained in the corresponding scenario keyword library. If they are inconsistent, the keywords are maintained in the keyword library that does not match the reported scenario.

[0045] (5) The alarm level is judged by the business alarm module after the aggregation results of machine learning and content recognition, and the alarm level and category are output. The business control module makes corresponding policy control based on the alarm level and category, and at the same time supplements and maintains the customer profile system.

[0046] This invention presents a privacy-preserving anti-fraud system and method based on machine learning big data models and Automatic Speech Recognition (ASR). Leveraging machine learning and big data models supported by indicator data, this dual approach significantly improves the efficiency of screening abnormal call groups and the accuracy of identifying and judging abnormal call content. Through preprocessing millions of call records, the machine learning module categorizes call groups based on their call characteristics, quickly identifying suspected fraudulent call groups. Keyword recognition and big data model judgment further verify the content of these suspected fraudulent call groups, enabling more accurate identification of fraudulent numbers and accelerating the control and response speed for abnormal call groups, thereby curbing fraud cases.

[0047] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the technical solutions of the present invention.

Claims

1. A privacy-preserving anti-fraud system for mobile accounts based on large-scale machine learning models and ASR (Automatic Recognition Scheme), characterized in that, include: The data statistics and preprocessing module is responsible for refining the basic call detail record data in the database, completing the entire data extraction, transformation, and loading process to achieve data preprocessing; Machine Learning Module: Responsible for performing cluster analysis on preprocessed data, predicting and classifying call groups based on key indicator data, and dividing calls into normal call groups and abnormal call groups; The automatic speech recognition module automatically recognizes the abnormal call groups predicted by the machine learning module and outputs the text results. The automatic inspection module has the function of creating scheduled tasks and can pull call recordings classified as fraudulent calls from the machine learning module to the automatic speech recognition module. The recognized text content is scheduled to keyword recognition and large model judgment tasks, so that the two models can perform tasks periodically and learn from the text content. The call content recognition and judgment module has keyword recognition and large model judgment functions. The keyword recognition function can identify the dialogue text content of abnormal call groups and judge whether there is high-risk fraud-related content in the dialogue. The large model judgment function can perform semantic understanding on the text results output by the automatic speech recognition module and judge whether it meets the characteristics of fraud-related calls. The business alarm module uses the process engine to aggregate machine learning data and content recognition results, assess the risk level and classification, and then push the alarm content to the business control module. The business management module responds with different management strategies based on the alarm level and alarm category of the business alarm module, while also supplementing and maintaining the customer profile system.

2. The privacy-preserving anti-fraud system for mobile accounts based on a large machine learning model and ASR as described in claim 1, characterized in that, The key metrics include call duration, connection rate, and percentage of short calls. Calls with a duration of less than 10 seconds, a connection rate of less than 40%, and a percentage of short calls exceeding 50% are classified as abnormal call groups.

3. The privacy-preserving anti-fraud system for mobile accounts based on a large machine learning model and ASR as described in claim 1, characterized in that, The risk levels are divided into Level 1 risk, Level 2 risk, and Level 3 risk. Level 1 risk is prohibited or violated regulations, Level 2 risk is suspected fraud, Level 3 risk is high-risk business, and Level 4 risk is malicious harassment.

4. The privacy-preserving anti-fraud system for mobile accounts based on a large machine learning model and ASR as described in claim 1, characterized in that, The control strategy includes: when the operations department receives alarms at each level, the business is assigned a priority for manual review according to the risk level, that is, the business of Level 1 and Level 2 risk is reviewed first. If the review results are correct, the business of Level 1 and Level 2 risk customers is directly frozen and taken offline. If the review results of Level 3 and Level 4 risk are correct, the business is taken offline as appropriate after communicating with the sales staff in charge.

5. A privacy-preserving anti-fraud method for mobile phone numbers based on large machine learning models and ASR, characterized in that, The anti-fraud system according to any one of claims 1 to 4 includes the following steps: (1) The basic call detail record information is extracted, transformed and loaded by the data statistics preprocessing module, and key call feature index data is output to the machine learning module. (2) The call feature index data output by the data preprocessing module is clustered by the machine learning module to divide the call groups into normal call groups and abnormal call groups. (3) The abnormal call group call recording files output by machine learning are recognized as text content through the automatic speech recognition module; (4) The keyword recognition function of the call content recognition and judgment module is used to identify keywords in the text content. If the text content matches the keyword, the matched keyword is extracted. If the text content does not match the keyword, the classification is manually reviewed to see if it is correct. If the classification is incorrect, the model is optimized. If the classification is correct, new keywords are labeled. The large model judgment function of the call content recognition and judgment module is used to judge whether the text content matches the abnormal call. If the text content matches the abnormal call, the judgment result is output. If the text content does not match the abnormal call, the judgment is manually reviewed to see if it is correct. If the judgment is incorrect, the model is optimized. Then the text learning results and machine learning results are aggregated. (5) The alarm level is judged by the business alarm module after the aggregation results of machine learning and content recognition, and the alarm level and category are output. The business control module makes corresponding policy control based on the alarm level and category, and at the same time supplements and maintains the customer profile system.

6. The anti-fraud method for private accounts based on a large machine learning model and ASR as described in claim 5, characterized in that, The model optimization method involves manually reviewing inaccurate call cases to determine whether the specific call scenario matches the actual reported scenario. If they match, keywords for that scenario are manually selected and the new keywords are added to the corresponding scenario keyword library. If they do not match, the keywords are added to the keyword library that does not match the reported scenario.

Images