Real-time voice phishing response apparatus based on large language model, method applied to the same
A real-time voice phishing response device using a large language learning model identifies and counters diverse voice phishing methods by converting call voices to text, applying TF-IDF and deep analysis, and executing predefined responses, thereby minimizing damage from voice phishing.
Patent Information
- Application Number
- KR1020240138473
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2024-10-11
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2044-10-11
AI Technical Summary
Existing technologies lack effective systemic solutions to identify and respond to the continuously diversifying methods of voice phishing, particularly over phone calls, which are difficult to detect due to manipulated caller IDs and often result in financial loss and identity theft.
A real-time voice phishing response device utilizing a large language learning model that converts call voices into text, applies TF-IDF and deep analysis to identify suspicious patterns, and executes predefined responses when risks are detected.
Minimizes the time required to identify and counter voice phishing, enabling real-time responses to prevent potential damage by continuously updating the model based on deep analysis and external data.
Smart Images

Figure 112024110576729-PAT00004_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a real-time voice phishing response device based on a large language learning model and a method applied thereto, and more specifically, to a real-time voice phishing response device based on a large language learning model capable of responding in real time to continuously diversifying voice phishing, and a method applied thereto. Background Technology
[0002] Traditional phishing methods practiced on the internet refer to a technique where a sender steals the website address (URL) of a bank or financial institution to collect personal information from a recipient. In this conventional phishing method, the sender sends a link via email containing a URL similar to that of a bank or financial institution (although there may be slight differences from the actual site, the recipient may perceive it as the same URL because they do not examine it closely). The recipient then clicks the link to access the site and enters personal information such as an ID, password, and verification code. After the sender obtains this personal information through a phishing server, they access the actual bank or financial institution site to cause financial damage or identity theft to the recipient.
[0003] In particular, phishing attacks using URLs similar to those of banks or financial institutions are becoming a social problem, as senders obtain financial information such as recipients' IDs, passwords, resident registration numbers, account numbers, and account passwords, causing significant damage to recipients including financial losses and identity theft.
[0004] As such damage on the internet has surged, security software designed to prevent phishing has emerged, raising awareness of the risks associated with accessing site information distributed via email or messengers.
[0005] Meanwhile, recently, telephone fraud crimes involving voice phishing using landlines or mobile phones, similar to the aforementioned phishing methods, are occurring frequently. However, since the recipient cannot identify the caller due to manipulation of the caller ID, there are cases where victims suffer damages helplessly.
[0006] For example, telephone financial fraud involves using landlines or mobile phones to impersonate banks, credit card companies, etc., to obtain financial information from the recipient, thereby causing damages such as account transfers, exposure of card numbers, and identity theft.
[0007] Furthermore, they obtain family information and the mobile phone numbers of victims or their children to impersonate kidnappers, causing psychological distress and financial fraud such as illegal bank transfers. They also impersonate the National Tax Service, the National Intelligence Service, or investigative agencies to induce recipients to leak personal information, resulting in identity theft or financial damage.
[0008] Since such voice phishing is generally carried out secretly by the caller manipulating the caller ID to transmit a random number, it is difficult for the recipient to determine the authenticity of the caller.
[0009] In addition, voice phishing is becoming more sophisticated, and its methods are expanding into more diverse fields and continuously changing. As a result, even if countermeasures against voice phishing are in place, victims of the changed voice phishing continue to suffer damages.
[0010] As mentioned above, there is still a lack of systemic solutions or technologies capable of identifying and responding to such voice phishing. Prior art literature
[0011] Korean Patent Publication No. 10-1338014 (December 2, 2013) The problem to be solved
[0012] Accordingly, the present invention was created to solve the above-mentioned problems, and the objective of the present invention is to provide a real-time voice phishing response device based on a massive language learning model capable of responding in real-time to continuously diversifying voice phishing, and a method applied thereto.
[0013] The purpose of the invention is not limited to the purposes mentioned above, and other unmentioned purposes will be clearly understood by those skilled in the art from the description below. means of solving the problem
[0014] A real-time voice phishing response device based on a big language learning model according to the first aspect of the present invention for achieving the above objective comprises a memory and at least one processor, wherein the processor converts a voice during a call into call text in real time, and for each of the one or more voice phishing processes corresponding to the call text, if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text is above a threshold value for each of the one or more voice phishing processes specified, it is determined as suspected voice phishing and provides guidance to the caller, and if the result of performing a deep voice phishing analysis based on a big language model on the call text is confirmed to be a voice phishing risk, one or more response measures for voice phishing predetermined are executed.
[0015] The above one or more voice phishing processes can be updated based on the results of the deep analysis of voice phishing based on the above big language model.
[0016] The parameters used to derive the above TF-IDF (Term Frequency-Inverse Document Frequency) analysis values may be changed according to the update of the corresponding voice phishing process among the above one or more voice phishing processes.
[0017] After processing the call text for network transmission according to a predetermined capacity, the result regarding the suspected voice phishing and the result regarding the voice phishing risk received as a response may be output, or the call text may be applied to a predetermined default voice phishing process, and if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value corresponding to the default voice phishing process is greater than or equal to a threshold value, the call text may be processed for network transmission to receive the response.
[0018] The deep analysis of voice phishing based on the above-mentioned big language model is executed in connection with an external API (Application Programming Interface) that provides the above-mentioned big language model, and the generated data for one or more voice phishing processes based on the results of executing the deep analysis of voice phishing based on the above-mentioned big language model, and the parameter data used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis values for each of the one or more voice phishing processes, may be stored in a database within a server that determines the suspicion of voice phishing and the risk of voice phishing.
[0019] And, a real-time voice phishing response method based on a big language learning model according to the second aspect of the present invention for achieving the above objective comprises, in a voice phishing response device, the step of converting a voice of a call in progress into call text in real time; the step of processing the call text to be transmitted to a network; the step of identifying one or more voice phishing processes corresponding to the call text, and then determining that the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text for each of the one or more identified voice phishing processes is above a threshold value, thereby determining it as suspected voice phishing and providing guidance to the caller; and the step of executing one or more predetermined response measures for voice phishing when the result of performing a voice phishing deep analysis based on a big language model on the call text is confirmed to be a voice phishing risk.
[0020] The above one or more voice phishing processes can be updated based on the results of the deep analysis of voice phishing based on the above big language model.
[0021] The parameters used to derive the above TF-IDF (Term Frequency-Inverse Document Frequency) analysis values may be changed according to the update of the corresponding voice phishing process among the above one or more voice phishing processes.
[0022] The method may further include the step of applying the above call text to a predetermined default voice phishing process and, if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value corresponding to the above default voice phishing process is greater than or equal to a threshold value, processing the transmission of the above call text to a network.
[0023] The deep analysis of voice phishing based on the above-mentioned big language model is executed in connection with an external API (Application Programming Interface) that provides the above-mentioned big language model, and the generated data for one or more voice phishing processes based on the results of executing the deep analysis of voice phishing based on the above-mentioned big language model, and the parameter data used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis values for each of the one or more voice phishing processes, may be stored in a database within a server that determines the suspicion of voice phishing and the risk of voice phishing. Effects of the invention
[0024] Therefore, the present invention has the advantage of minimizing the time required to identify and prepare countermeasures against continuously diversifying voice phishing, and thereby supporting real-time response during a call to prevent damage to potential victims of voice phishing based on continuously diversifying voice phishing.
[0025] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description in the claims. Brief explanation of the drawing
[0026] FIG. 1 is a configuration diagram showing a voice phishing response device according to one embodiment of the present invention. FIG. 2 is a configuration diagram showing a voice phishing response device according to another embodiment of the present invention. Figure 3 is an example diagram illustrating the first stage execution of the voice phishing response device and server of Figure 1. Figure 4 is an example diagram illustrating the two-stage execution of the voice phishing response device and server of Figure 1. And, FIG. 5 is a flowchart illustrating a voice phishing response method according to one embodiment of the present invention. Specific details for implementing the invention
[0027] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.
[0028] Furthermore, the embodiments described herein will be explained with reference to cross-sectional views and / or schematic drawings, which are exemplary illustrations of the present invention. Accordingly, the form of the exemplary drawings may be modified due to manufacturing techniques and / or tolerances, etc. Additionally, in each drawing of the present invention, each component may be depicted slightly enlarged or reduced for convenience of explanation.
[0029] The real-time voice phishing response device based on the big language learning model of the present invention, and the method applied thereto, are configured to minimize the time required to identify and prepare response measures for continuously diversifying voice phishing, and thereby enable real-time response during a call to block damage to potential victims of voice phishing based on continuously diversifying voice phishing.
[0030] FIG. 1 is a configuration diagram showing a voice phishing response device according to one embodiment of the present invention.
[0031] As illustrated in FIG. 1, a real-time voice phishing response device (200) based on a large language learning model includes a memory (210) and at least one processor (220). The processor (220) converts the voice of a call being made into call text in real time. After identifying one or more voice phishing processes corresponding to the converted call text, if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text for each identified one or more voice phishing processes is above a threshold value, it is determined to be suspected voice phishing and the caller is notified. When the result of performing a deep voice phishing analysis based on a large language model on the call text confirms that there is a voice phishing risk, one or more predetermined response measures for voice phishing are executed.
[0032] Here, one or more voice phishing processes may include voice phishing fields (e.g., impersonation of a prosecutor, kidnapping of a child, consent for emergency surgery due to an accident involving a family member, financial loans, etc.), voice phishing development patterns (e.g., in the case of child kidnapping, 1) notifying of the kidnapping by stating the child's name → 2) threatening to harm the child if the kidnapping is reported to the police or other external parties → 3) inducing an immediate decision without giving time → 4) demanding money such as virtual currency or a transfer to a specific account), and keywords for each stage of the process, and refers to a process that supports the rapid identification of diversifying voice phishing and enables response during a voice phishing call.
[0033] In the case of child abduction, cited as an example of the aforementioned voice phishing development patterns, it is a commonly used voice phishing pattern, and when compared to actual abduction, there are the following major differences.
[0034] Differences in request methods
[0035] Voice Phishing: Criminals demand urgent action and typically request immediate money transfers. In such cases, the methods used to demand money often center on non-face-to-face transactions, such as bank transfers, virtual currencies, or Bitcoin.
[0036] Actual Kidnapping: In actual kidnappings, physical evidence of the child (e.g., belongings, photos, etc.) is often presented in addition to receiving money; while there is a demand for money, the focus is often on the reality of the crime and the threat involved.
[0037] Psychological pressure and time
[0038] Voice phishing: Criminals demand urgent action and require an immediate response without giving the victim time. They often tend to keep the call going to prevent the victim from contacting the police or checking with their children.
[0039] Actual Kidnapping: In the case of actual kidnapping, kidnappers have a strong tendency to give time for negotiation. In most cases, they continue negotiations based on the victim's reaction, allowing the victim time to grasp the facts of the kidnapping.
[0040] Therefore, it is possible to establish a voice phishing process for child abduction based on the aforementioned key differences.
[0041] Furthermore, if the aforementioned voice phishing regarding child abduction becomes more sophisticated and presents a "photo of a kidnapped child" as physical evidence by hacking the child's photo and forging it into a "photo of a kidnapped child" through AI image generation, it is possible to additionally generate a voice phishing process related to child abduction that reflects the "case of presenting a forged photo of a kidnapped child."
[0042] Since the likelihood of falling victim to voice phishing increases when a 'photo of a child who is being held captive' is presented, guiding the caller on the possibility of whether it is 'voice phishing (determining whether the photo of the kidnapped child is forged during the call, or determining it as voice phishing if the weight is high enough to demand a money transfer while on the call without giving time for negotiation even if the photo of the child is provided)' or 'actual kidnapping' through a voice phishing process that reflects this is meaningful in minimizing damage from voice phishing.
[0043] Therefore, it is desirable that the voice phishing process described above be continuously updated based on the results of deep analysis of voice phishing based on a big language model or data collected by the big language model itself.
[0044] In addition, a large language model typically refers to one having hundreds of millions to billions of parameters, and extending this further, it refers to a model having billions to hundreds of billions of parameters or a model having more parameters. For example, GPT-2, which has 1.5 billion parameters, corresponds to a typical large language model, and GPT-3, which has 175 billion parameters, and GPT-4, which has even more parameters, can also correspond to the large language model of the present invention.
[0045] TF-IDF (Term Frequency-Inverse Document Frequency) analysis values for call text can be performed for one or more voice phishing processes, and for example, if a voice phishing process corresponding to the aforementioned child abduction is identified, TF-IDF (Term Frequency-Inverse Document Frequency) analysis can be performed on the call text based on the voice phishing process of the identified child abduction.
[0046] TF-IDF (Term Frequency-Inverse Document Frequency) analysis analyzes statistical figures indicating how important specific words are in call text, and can be analyzed using the following example.
[0047] TF (Term Frequency) is defined as a value that measures how often a specific word appears in a text message. It can be measured by dividing the number of times a specific word appears in a text message by the total number of words, as shown in Formula 1 below.
[0048] <Formula 1>
[0049] TF(t) = Number of times a specific word appeared in the call text / Total number of words in the call text
[0051] IDF (Inverse Document Frequency) is defined as a value that measures how rare a specific word is in the entire set of call texts. It can be calculated inversely proportional to the number of call texts in which a specific word appears (the number of documents in multiple call texts for various call cases). In other words, a low weight can be assigned to a word that appears frequently in call texts in other cases as well as in the current call, and a relatively high weight can be assigned to a word that appears infrequently in some call texts. It can be measured as shown in Equation 2 below.
[0052] <Equation 2>
[0053]
[0055] In addition, when calculating the IDF value, if a specific word does not appear in the entire document, a division by zero error may occur, so it is also possible to calculate by adding 1 as in Formula 3 below.
[0056] <Equation 3>
[0057]
[0058] TF-IDF (Term Frequency-Inverse Document Frequency) is calculated by multiplying TF and IDF values as shown in Equation 4 below to indicate the relative importance of specific words within a call text; words that appear frequently in the call text but rarely in other documents are considered to have a high weight. However, the degree of weight and the criteria for reflection may vary depending on the characteristics of the voice phishing process.
[0059] <Equation 4>
[0060]
[0062] As such, the parameters used to calculate the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value (e.g., specific words described earlier) may be changed or added depending on updates to the voice phishing process.
[0063] Such processing can be performed by a voice phishing response device (200), such as a smartphone, after it processes the transmission of a call text over a network, and then by a server (300) that receives it. At this time, the voice phishing response device (200) can process the transmission of the call text based on time or by processing the transmission according to a predetermined capacity.
[0064] Additionally, the deep voice phishing analysis based on a large language model executed on the server (300) is executed in connection with an external API (Application Programming Interface) that provides the large language model, and the parameter data used to derive the generated data for one or more voice phishing processes described above and the F-IDF (Term Frequency-Inverse Document Frequency) analysis values for each of the one or more voice phishing processes based on the results of the deep voice phishing analysis based on the large language model can be stored in the server (300) that determines voice phishing suspicion and voice phishing risk, or in a database linked to the server (300).
[0065] FIG. 2 is a configuration diagram showing a voice phishing response device according to another embodiment of the present invention.
[0066] As illustrated in FIG. 2, the voice phishing response device (400) may apply a predetermined default voice phishing process before processing the transmission of a call text to a network, and if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value corresponding to the applied default voice phishing process is greater than or equal to a threshold value, the call text may be processed for transmission to a network.
[0067] Here, the threshold value refers to a reference value for determining whether the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value is at a level where damage may occur due to voice phishing, and it can be specified by reflecting the value measured during the actual implementation process and may vary depending on the voice phishing process. That is, the threshold value can be interpreted as having the same meaning as the threshold value described in Fig. 1, as well as the default voice phishing process described in Fig. 2.
[0068] Figure 3 is an example diagram illustrating the first stage execution of the voice phishing response device and server of Figure 1.
[0069] As illustrated in figure 3a), the server (300) identifies one or more voice phishing processes corresponding to the received call text, and for each of the identified one or more voice phishing processes, if a TF-IDF (Term Frequency-Inverse Document Frequency) analysis is performed in the first step and it is determined to be suspected of voice phishing, the server can provide the corresponding result to the voice phishing response device (200).
[0070] Accordingly, the voice phishing response device (200) can process a notice on the call screen during a call, as shown in Figure 3b), meaning "there is a possibility of voice phishing, so please wait for the analysis results," as an example of a response measure.
[0071] Figure 4 is an example diagram illustrating the two-stage execution of the voice phishing response device and server of Figure 1.
[0072] As illustrated in figure 4a), if the server (300) is classified as 'suspected voice phishing' in the first stage analysis of the received call text, it performs a voice phishing deep analysis based on a large language model, and if the result is determined to be 'voice phishing risk', it can provide the corresponding result to the voice phishing response device (200).
[0073] Accordingly, the voice phishing response device (200) can process a notice meaning 'voice phishing risk' on the call screen while in a call as an example of a response measure, as shown in Figure 4b).
[0074] In this regard, the server (300) may also take active measures, such as notifying relevant agencies (e.g., police investigation authorities) of information regarding a voice phishing suspected device (e.g., smartphone, 200) that is in contact with a voice phishing response device (e.g., smartphone, 200) and related call text history, when there is a ‘voice phishing risk’.
[0075] And, FIG. 5 is a flowchart illustrating a voice phishing response method according to one embodiment of the present invention.
[0076] As illustrated in FIG. 5, a real-time voice phishing response method based on a large language learning model proceeds by converting the voice of a call being made into call text in real time at a voice phishing response device (200 or 400) (S100).
[0077] In step S100, the converted call text is processed for transmission to the network (S102).
[0078] In step S102, the call text processed for network transmission is processed by a server for processing it, and after identifying one or more voice phishing processes corresponding to the call text, if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text for each identified one or more voice phishing processes is above a threshold value, it is determined to be suspected of voice phishing and the determination result is provided to the aforementioned voice phishing response device, and the voice phishing response device (200 or 400) receives this and provides guidance to the caller through a screen, etc. (S104 to S108).
[0079] Additionally, if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text is greater than or equal to a threshold value, the server (300) performs a deep voice phishing analysis based on a large language model and additionally provides the result to the voice phishing response device described above, and the voice phishing response device (200 or 400) receives this and guides the caller to take response measures through a screen, etc. (S110 to S114).
[0080] The detailed description of steps S100 to S114 described above and the description of additional possible steps shall be in accordance with FIGS. 1 to 4 and the detailed description thereof.
[0081] Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without changing its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. Industrial applicability
[0082] Furthermore, since the present invention aims to provide a real-time voice phishing response device based on a massive language learning model capable of responding in real time to continuously diversifying voice phishing, and a method applied thereto, it is an invention with industrial applicability as it not only has sufficient potential for commercialization or business but is also clearly and practically implementable. Explanation of the symbols
[0083] 100: Suspected voice phishing device 200, 400: Voice phishing response device 210, 410 : Memory 220, 420 : Processor 300 : Server
Claims
Claim 1 Memory; and includes at least one processor, wherein the processor converts the voice of a call in progress into call text in real time, applies the call text to a predetermined default voice phishing process, and if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text corresponding to the default voice phishing process is above a threshold, processes the call text for transmission to a network, and after identifying one or more voice phishing processes corresponding to the call text transmitted to the network, if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text for each identified one or more voice phishing process is above a threshold, it is determined as suspected voice phishing and a notification is processed to the caller, wherein the one or more voice phishing processes include at least one of a voice phishing field, a voice phishing deployment pattern, and keywords for each stage of the process, and the deep voice phishing analysis based on a mega-language model for the call text is executed in connection with an external API (Application Programming Interface) that provides the mega-language model, and the result of executing the deep voice phishing analysis based on a mega-language model for the call text is voice If a phishing risk is identified, one or more predetermined response measures for voice phishing are executed, the one or more voice phishing processes are updated based on the results of the deep voice phishing analysis based on the big language model, and the parameters used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value are changed according to the update of the corresponding voice phishing process among the one or more voice phishing processes.A real-time voice phishing response device based on a massive language learning model, wherein the generated data for one or more voice phishing processes based on the results of performing deep voice phishing analysis based on the above massive language model, and the parameter data used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis values for each of the above one or more voice phishing processes, are stored in a database within a server that identifies the suspected voice phishing and the voice phishing risk. Claim 2 In claim 1, the one or more voice phishing processes are a real-time voice phishing response device based on a massive language learning model that is updated based on the results of the deep voice phishing analysis based on the massive language model. Claim 3 In claim 2, the parameter used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value is a real-time voice phishing response device based on a massive language learning model that changes according to the update of a corresponding voice phishing process among the one or more voice phishing processes. Claim 4 A real-time voice phishing response device based on a massive language learning model according to claim 1, wherein the device processes the call text for network transmission according to a predetermined capacity and outputs the result regarding the suspected voice phishing and the result regarding the risk regarding voice phishing received as a response, or applies the call text to a predetermined default voice phishing process and processes the call text for network transmission and receives the response when the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value corresponding to the default voice phishing process is greater than or equal to a threshold value. Claim 5 In claim 1, the deep voice phishing analysis based on the big language model is executed in connection with an external API (Application Programming Interface) that provides the big language model, and the parameter data used to derive the generated data for one or more voice phishing processes based on the result of executing the deep voice phishing analysis based on the big language model and the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value for each of the one or more voice phishing processes is a real-time voice phishing response device based on a big language learning model that is stored in a database within a server that determines the suspected voice phishing and the voice phishing risk. Claim 6 A voice phishing response device comprises: a step of converting a voice of a call in progress into call text in real time; a step of applying the call text to a predetermined default voice phishing process and processing the transmission of the call text to a network if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text corresponding to the default voice phishing process is above a threshold; a step of identifying one or more voice phishing processes corresponding to the call text transmitted to the network, and if the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value of the call text for each identified one or more voice phishing processes is above a threshold, determining it as suspected voice phishing and providing guidance to the caller; and a step of executing a voice phishing deep analysis based on a massive language model for the call text by connecting it to an external API (Application Programming Interface) that provides the massive language model.The method includes a step of executing one or more predetermined countermeasures against voice phishing when the result of performing a deep voice phishing analysis based on a big language model on the above call text confirms a voice phishing risk; wherein the one or more voice phishing processes include at least one of a voice phishing field, a voice phishing development pattern, and keywords for each stage of the process; wherein the one or more voice phishing processes are updated based on the results of the deep voice phishing analysis based on the big language model; wherein the parameters used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value are changed according to the update of the corresponding voice phishing process among the one or more voice phishing processes; and wherein the generated data for the one or more voice phishing processes based on the results of the deep voice phishing analysis based on the big language model and the parameter data used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value for each of the one or more voice phishing processes are based on a big language learning model stored in a database within a server that determines the voice phishing suspicion and the voice phishing risk. A real-time voice phishing response method. Claim 7 In claim 6, the above one or more voice phishing processes is a real-time voice phishing response method based on a massive language learning model that is updated based on the results of the deep analysis of voice phishing based on the massive language model. Claim 8 In claim 7, the parameter used to derive the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value is a real-time voice phishing response method based on a massive language learning model that changes according to the update of a corresponding voice phishing process among the one or more voice phishing processes. Claim 9 A real-time voice phishing response method based on a massive language learning model according to claim 6, further comprising the step of applying the call text to a predetermined default voice phishing process and processing the transmission of the call text to a network when the TF-IDF (Term Frequency-Inverse Document Frequency) analysis value corresponding to the default voice phishing process is greater than or equal to a threshold value. Claim 10 A real-time voice phishing response method based on a massive language learning model according to claim 6 or 9, wherein the deep voice phishing analysis based on the massive language model is executed in connection with an external API (Application Programming Interface) providing the massive language model, and the parameter data used to derive the generated data for the one or more voice phishing processes and the TF-IDF (Term Frequency-Inverse Document Frequency) analysis values for each of the one or more voice phishing processes based on the results of executing the deep voice phishing analysis based on the massive language model is stored in a database within a server that determines the suspected voice phishing and the voice phishing risk.
Citation Information
Patent Citations
Apparatus and method for analyzing voice phishing pattern based on probability
KR101792204B1
Technology for phishing awareness and phishing detection
US20240333750A1
Apparatus for Preventing Phishing Fraud and Driving Method Thereof
KR102707506B1
Detecting phishing attempts
US20200067978A1
