System

Abstract

A system is provided.SOLUTION: A system including: means for collecting voice input; means for converting the collected voice input into text; means for detecting a specific keyword in the text; means for analyzing the context before and after the detected keyword to evaluate a possibility of fraud; means for generating a question for user authentication when the possibility of fraud is high; means for displaying the generated question; means for collecting a user's answer; and means for evaluating the collected answer and determining whether the user is legitimate.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The technology of the present disclosure relates to a system. [Background technology]

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] In today's world, fraudulent activities using telephones and the Internet are on the rise, making the elderly and those less familiar with technology particularly vulnerable to fraud. Conventional fraud prevention measures lack systems that can detect signs of fraud before a fraud is committed and take appropriate countermeasures. Therefore, there is a need for a system that can detect fraud and prevent it quickly and effectively. [Means for solving the problem]

[0005] The present invention provides a means for first collecting voice input and converting it to text. It then provides a means for detecting specific keywords in the converted text, analyzing the context surrounding the detected keywords, and assessing the likelihood of fraud. Furthermore, if there is a high likelihood of fraud, it generates user authentication questions and displays the questions to the user. It also provides a means for collecting answers from the user and evaluating the collected answers to determine whether the user is legitimate. By combining these means, a system that effectively detects and prevents fraud is realized.

[0006] "Voice input" refers to operations or instructions given by the user through voice.

[0007] "Means of collection" refers to devices or software that have the ability to record voice input and save it as data.

[0008] "Means for converting to text" refers to the technology or algorithms used to analyze and convert audio data into natural language text format.

[0009] "Specific keywords" refer to words or phrases that are important in determining whether a transaction is potentially fraudulent.

[0010] "Means of detection" refers to the technology or algorithms that search for and extract specific keywords within text data.

[0011] "Contextual analysis methods" refers to technologies and algorithms that understand the context of text and evaluate the meaning and relevance of specific keywords.

[0012] "Means for assessing the likelihood of fraud" refers to technologies or algorithms that quantify or score the risk of fraud based on the results of contextual analysis.

[0013] "Means for generating user authentication challenges" refers to a technology or system that generates personalized challenges for the purpose of verifying a user's authenticity in cases of potential fraud.

[0014] "Means for displaying a question" refers to a device or software that presents the generated question to the user visually or audibly.

[0015] "Means for collecting user answers" refers to any device or software capable of receiving data input from a user to answer an authentication question.

[0016] "Means for evaluating answers" refers to the technology or algorithm that verifies whether a user's answer is correct and determines its validity.

[0017] "System" refers to the set of technologies and devices that integrate these measures to operate effectively and detect and prevent fraud. [Brief explanation of the drawings]

[0018] [Figure 1] 1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6] FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9]1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11] FIG. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0019] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0020] First, the terms used in the following description will be explained.

[0021] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, a processor may be a single type of arithmetic device or a combination of multiple types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), and an APU (Accelerated Processing Unit).

[0022] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0023] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0024] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0025] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0026] [First embodiment]

[0027] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0028] 1, a data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0029] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0030] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0031] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0032] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0033] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0034] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0035] 2, in the data processing device 12, a specific process is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific process is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0036] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0037] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0038] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0039] This invention relates to a system that detects potential fraudulent activity based on voice input and, if necessary, authenticates the user. This system is realized by combining speech recognition technology and natural language processing (NLP).

[0040] System configuration and operation

[0041] 1. Collecting and converting voice input

[0042] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[0043] 2. Keyword detection and context analysis

[0044] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[0045] 3. Generating and Displaying Authentication Questions

[0046] If fraud is deemed likely, the server generates a challenge, which is tailored to the individual user, such as information about the user's relatives, and is sent to the user's device for display.

[0047] 4. Collecting and Evaluating User Responses

[0048] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[0049] Specific examples

[0050] The specific flow when the user says "transfer money" is shown below.

[0051] 1. User: Say "transfer money" into the terminal.

[0052] 2. Device: Records audio and sends the data to the server.

[0053] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[0054] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[0055] 5. Server: Generate the authentication question "What is your mother's maiden name?"

[0056] 6. Server: Sends the question to the user's terminal and displays it.

[0057] 7. User: Answers "Tanaka."

[0058] 8. Terminal: Send this response to the server.

[0059] 9. Server: Evaluate whether "Tanaka" is legitimate.

[0060] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[0061] In this way, the system analyzes user statements through multiple steps to prevent fraud.

[0062] The processing flow will be explained below.

[0063] Step 1:

[0064] The user makes a statement such as "transfer money" through voice input.

[0065] Step 2:

[0066] The terminal records the user's voice in real time and saves it as voice data.

[0067] Step 3:

[0068] The terminal transmits the recorded voice data to the server.

[0069] Step 4:

[0070] The server converts the received voice data into text using a voice recognition module.

[0071] Step 5:

[0072] The server detects specific keywords (e.g., "transfer money") in the converted text.

[0073] Step 6:

[0074] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[0075] Step 7:

[0076] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[0077] Step 8:

[0078] The server generates a user authentication question, such as "What is your mother's maiden name?"

[0079] Step 9:

[0080] The server sends the generated authentication challenge to the user's terminal.

[0081] Step 10:

[0082] The terminal displays the authentication challenge received from the server to the user.

[0083] Step 11:

[0084] The user answers the displayed authentication question (e.g., answers "Tanaka").

[0085] Step 12:

[0086] The terminal transmits the user's answer to the server.

[0087] Step 13:

[0088] The server evaluates the received user answers and verifies their validity.

[0089] Step 14:

[0090] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[0091] Example 1

[0092] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0093] While user interfaces using voice input have become widespread in recent years, fraudulent acts by malicious third parties have also increased. It is particularly important to prevent fraudulent acts when transactions are made using voice input, such as when exchanging personal information. However, conventional voice recognition systems have not been able to adequately detect fraudulent acts and authenticate users, and a highly reliable authentication system is needed.

[0094] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0095] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating user authentication questions if the possibility of fraud is high, means for generating user authentication questions using a generative AI model, means for transmitting the generated questions to a user terminal and displaying them, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, and means for notifying the user terminal of the authentication result. This enables fraud detection and prevention through advanced voice input analysis and authentication processes.

[0096] "Voice input" refers to recording what the user says as voice data on the terminal.

[0097] "Voice data" refers to the digital signal obtained from voice input, and is data that is sent to a server for analysis.

[0098] "Text data" refers to character string information converted from voice data using voice recognition technology.

[0099] A "keyword" is a key word or phrase extracted from voice input or text data based on specific conditions.

[0100] "Contextual analysis" is the process of evaluating the context within text data and analyzing its overall meaning and relevance.

[0101] "Possibility of fraud" refers to a score or probability that evaluates the likelihood that the content of voice input or text data constitutes fraud.

[0102] An "authentication challenge" is a question generated to verify the authenticity of a user.

[0103] A "generative AI model" is an artificial intelligence that uses machine learning techniques to generate new questions and text.

[0104] A "user terminal" is a device that allows a user to input voice and receive authentication questions, and includes a smartphone, computer, etc.

[0105] "User authentication" is the process of determining whether a user is legitimate based on collected information (e.g., answers to authentication questions).

[0106] The "authentication result" is the result of user authentication, and is information indicating whether the authentication was successful or unsuccessful.

[0107] MODE FOR CARRYING OUT THE INVENTION

[0108] This invention relates to a system that detects possible fraudulent activity based on voice input and authenticates the user as necessary. This system is realized by combining voice recognition technology and natural language processing (NLP). Specific embodiments for carrying out the invention are described below.

[0109] System configuration

[0110] This system mainly consists of a server, a terminal, and a user. Each component functions as follows:

[0111] Collecting and sending voice input

[0112] The user speaks to the terminal, saying something like "transfer money." The terminal records the voice using a built-in microphone and generates voice data. This terminal can be a smartphone, tablet, or PC. The recorded voice data is sent to the server using HTTPS as the communication protocol.

[0113] Audio data conversion

[0114] The server converts the received voice data into text data using a voice recognition module such as Google® Cloud Speech-to-Text API. For example, voice data uttered "Transfer money" is converted into text data stating "Transfer money."

[0115] Keyword detection and context analysis

[0116] The server receives the text data and uses the natural language processing library spaCy to detect specific keywords within the text. If a keyword is detected, the server analyzes the context and assesses the likelihood of fraud by assigning a score ranging from 0% to 100%.

[0117] Generate and submit authentication questions

[0118] If it is determined that there is a high possibility of fraudulent activity, the server uses a generative AI model (e.g., GPT-4 (registered trademark)) to generate a question for user authentication. For example, a question such as "What is your mother's maiden name?" may be generated. This question includes information about the user's close relationships. The generated question is sent to the user's device and displayed on the device screen.

[0119] Collecting and evaluating user responses

[0120] The user enters an answer to the displayed authentication question. The device collects this answer and sends it to the server. The server receives the user's answer and compares it with a pre-configured correct answer database to evaluate whether it is valid. For example, if the user answers "Tanaka," it verifies whether the answer is valid.

[0121] Notification of authentication results

[0122] The server determines whether the authentication was successful based on the user's answers and notifies the user terminal of the result. If the authentication was successful, a message to that effect is displayed on the terminal screen.

[0123] Specific examples

[0124] A specific example in which the user says "transfer money" will be shown below.

[0125] 1. User: Say "transfer money" into the terminal.

[0126] 2. Device: Records audio and sends the data to the server.

[0127] 3. Server: Convert the audio data into text using Google Cloud Speech-to-Text. Detect the phrase "transfer money."

[0128] 4. Server: Uses spaCy to analyze the context and scores the likelihood of fraud at 80%.

[0129] 5. Server: Use GPT-4 to generate the authentication question "What is your mother's maiden name?"

[0130] 6. Server: Sends the question to the user's terminal and displays it.

[0131] 7. User: Answers "Tanaka."

[0132] 8. Terminal: Send this response to the server.

[0133] 9. Server: Evaluate whether "Tanaka" is legitimate.

[0134] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[0135] In this way, the system prevents fraud through advanced voice input analysis and authentication processes.

[0136] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0137] Step 1:

[0138] User: The user speaks into the terminal, saying "transfer money."

[0139] Specific action: Record audio using a microphone.

[0140] Input: User's voice

[0141] Output: Recorded audio data

[0142] Step 2:

[0143] Device: The device collects the recorded voice data and sends it to the server.

[0144] Specific operation: The recorded audio data is sent to the server via HTTPS.

[0145] Input: Recorded audio data

[0146] Output: Audio data sent to the server

[0147] Step 3:

[0148] Server: The server converts the received voice data into text data using a speech recognition module (e.g., Google Cloud Speech-to-Text).

[0149] Specific operation: Sends audio data to the API and receives it as text data.

[0150] Input: Audio data

[0151] Output: Text data

[0152] Step 4:

[0153] Server: The server uses a natural language processing library (e.g., spaCy) to detect whether specific keywords (e.g., "transfer money") are contained in the text data.

[0154] Specific operation: Matches text data with a keyword list and extracts matching keywords.

[0155] Input: Text data

[0156] Output: Text data containing keywords

[0157] Step 5:

[0158] Server: If a keyword is detected, the context is analyzed and a score is assigned for potential fraud.

[0159] What it does: Runs a contextual analysis algorithm and scores the likelihood of fraud on a scale of 0% to 100%.

[0160] Input: Text data containing keywords

[0161] Output: A score indicating the likelihood of fraud.

[0162] Step 6:

[0163] Server: If fraud is deemed likely, a generative AI model (e.g., GPT-4) is used to generate user authentication questions.

[0164] Specific operation: The AI ​​model is given a prompt, "Generate a question using information about the user's close relationships," and the generated question is received.

[0165] Input: Fraud likelihood score, prompt statement

[0166] Output: Generated authentication questions

[0167] Step 7:

[0168] Server: Sends the generated authentication question to the user's terminal and instructs it to be displayed.

[0169] Specific operation: The text data of the question is sent to the terminal via HTTPS.

[0170] Input: Generated authentication question

[0171] Output: Authentication question displayed on the user's terminal

[0172] Step 8:

[0173] User: The user enters the answer to the displayed authentication question. For example, the user answers "Tanaka."

[0174] Specific actions: Enter your answer in the input field on the device and press the send button.

[0175] Input: Answer to authentication question

[0176] Output: Answers typed into the terminal

[0177] Step 9:

[0178] Terminal: The terminal collects the user's answers and sends them to the server.

[0179] Specific operation: The entered answer is sent to the server via HTTPS.

[0180] Input: Answer typed into the terminal

[0181] Output: The answer sent to the server

[0182] Step 10:

[0183] Server: Compares the received user answers with a pre-configured database of correct answers and evaluates whether the user is legitimate.

[0184] Specific behavior: Compares the answer data with the correct answer database to see if it matches.

[0185] Input: User answers, correct answer database

[0186] Output: Evaluation result (valid or not)

[0187] Step 11:

[0188] Server: Generates the authentication result and notifies the user terminal.

[0189] Specific operation: Creates a message indicating whether authentication was successful or failed and sends it to the terminal via HTTPS communication.

[0190] Input: Evaluation result

[0191] Output: Authentication result sent to the terminal

[0192] Step 12:

[0193] Terminal: Displays the authentication result received from the server to the user.

[0194] Specific behavior: A message indicating authentication success or failure is displayed on the screen.

[0195] Input: Authentication result

[0196] Output: Authentication result displayed to the user

[0197] (Application example 1)

[0198] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0199] Current voice input systems have difficulty detecting fraud in real time and quickly authenticating users. Especially in electronic payments, where transactions are frequent and fast, there is a lack of means to prevent fraud immediately. Furthermore, fraud detection systems based on voice input often rely too much on specific keywords, making it difficult to accurately analyze the context, leading to false positives. Therefore, a system that can detect fraud and authenticate users with higher accuracy is needed.

[0200] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0201] In this invention, the server includes means for collecting voice input in real time, means for converting the voice input into text, means for detecting specific keywords in the text, means for analyzing the context and assessing the likelihood of fraud, means for generating additional authentication challenges if the likelihood of fraud is high, and means for displaying the generated challenges on the user terminal, thereby enabling instant fraud detection using voice and rapid user authentication.

[0202] "Voice input" refers to spoken content or instructions from a user through a voice collection device such as a microphone.

[0203] "Means for converting to text" refers to speech recognition technology or software for converting voice input into character string data.

[0204] "Means for detecting specific keywords" refers to the ability to identify and extract predefined important words and phrases within the converted text.

[0205] "Contextual fraud risk assessment" refers to algorithms that analyze the context of text and the use of language to score or assess the risk of fraud.

[0206] The "means for generating additional authentication questions" refers to the ability to create new questions to verify the user's authenticity if it is determined that there is a high possibility of fraud.

[0207] "Means for displaying on the user terminal" refers to the technology or interface for displaying the generated authentication question on the screen of the device used by the user.

[0208] "Means for recognizing voice in real time" refers to a function that instantly detects and recognizes the user's speech and then processes it.

[0209] "Means for obtaining analysis results on the server" refers to the function of obtaining the results of speech recognition and text analysis from a remote server and sharing and using them throughout the system.

[0210] A "generative AI model" is a machine learning model that uses artificial intelligence to analyze and process input data and generate appropriate output.

[0211] A "prompt" is a document containing instructions or questions to be input into a generative AI model, providing criteria and conditions for data analysis.

[0212] DETAILED DESCRIPTION OF THE INVENTION The following describes an embodiment of the present invention.

[0213] The server has a means for collecting voice input, and collects the voice spoken by the user into the terminal. The collected voice is converted into text using voice recognition technology. An example of the voice recognition technology used here is Google's voice recognition API. By using this API, the collected voice data can be converted into text in real time.

[0214] The converted text is sent to a server where it is processed to detect specific keywords. Natural Language Processing (NLP) techniques are recommended to detect specific keywords in the text. Once a specific keyword is detected, the server analyzes the context and assesses its likelihood of fraud. This assessment is performed using a generative AI model, which analyzes the prompt sentence as input.

[0215] If the server determines that fraud is likely, it generates additional questions for user authentication. These questions may contain information about the user's relatives, for example, and are important for verifying the user's authenticity. The generated questions are sent to the user's terminal and displayed on its screen.

[0216] The user answers the displayed questions, and the answers are sent to the server via the device. The server evaluates the user's answers and determines whether the user is legitimate. This authentication uses a generative AI model to instantly evaluate the validity of the answers.

[0217] Hardware or software used

[0218] Hardware: Smartphone, microphone

[0219] software:

[0220] Speech Recognition API (Google)

[0221] Natural Language Processing (NLP)

[0222] Generative AI Models

[0223] Libraries for handling HTTP requests (requests)

[0224] Realization of processing steps

[0225] For example, if the user says "I would like to purchase this item," the following steps are performed:

[0226] 1. Audio collection and conversion:

[0227] The user speaks into their smartphone, saying, "I would like to purchase this product." The device collects this speech and converts it into text using a speech recognition API.

[0228] 2. Text transmission and analysis:

[0229] The text is sent to a server, which detects specific keywords (such as "purchase") and analyzes the context. If it is determined to be fraudulent, the corresponding prompt is fed into a generative AI model.

[0230] 3. Generate and display authentication questions:

[0231] If the generative AI model scores a high probability of fraud, it generates additional authentication questions and displays them on the user's device, such as "What is your mother's maiden name?"

[0232] 4. User answers and their ratings:

[0233] If the user answers "Tanaka," the device sends the answer to the server, which evaluates the answer using a generative AI model to confirm its validity.

[0234] Prompt Sentence Examples

[0235] An example of a prompt used in this system is as follows:

[0236] "Please enter the following information into the system:

[0237] Say: "Buy this item."

[0238] Contextual analysis results: 80% probability of fraud.

[0239] Verification Question: What is your mother's maiden name?

[0240] User Answer: Tanaka.

[0241] Please evaluate the user for legitimacy and notify us of the results."

[0242] This embodiment allows for instant fraud detection and highly accurate user authentication based on voice input.

[0243] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0244] Step 1:

[0245] A user speaks into a device (smartphone) to input information. For example, the user might say, "I would like to purchase this product."

[0246] Input: User's voice data

[0247] Output: Collected audio data

[0248] Step 2:

[0249] The device sends the collected voice data in real time to a voice recognition API (for example, Google's voice recognition API), which converts the voice data into text data.

[0250] Input: Collected audio data

[0251] Output: Converted text data

[0252] Specific operation: Sends audio data to the API and receives text data from the API.

[0253] Step 3:

[0254] The device transmits the converted text data to a server, which performs text analysis to detect specific keywords in the text data, for example, the keyword "purchase."

[0255] Input: Text data

[0256] Output: Keyword detection results

[0257] What it does: Runs a text analysis algorithm to extract specific keywords.

[0258] Step 4:

[0259] The server analyzes the context of the detected keywords and utilizes a generative AI model to assess the likelihood of fraud. The prompt sentence is provided as input to the generative AI model, which scores the probability of fraud.

[0260] Input: Text data and detected keywords

[0261] Output: Fraud likelihood score

[0262] Specific operation: Contextual analysis is performed using NLP technology, prompt sentences are input into the generative AI model, and a score is calculated.

[0263] Step 5:

[0264] If the fraud score is high, the server generates additional questions to authenticate the user, for example, "What is your mother's maiden name?"

[0265] Input: Fraud likelihood score

[0266] Output: Generated authentication questions

[0267] What it does: Selects and generates appropriate questions from a generative AI model or a predefined question set.

[0268] Step 6:

[0269] The generated user authentication question is sent from the server to the terminal and displayed on the terminal screen.

[0270] Input: Generated authentication question

[0271] Output: Question displayed on terminal

[0272] Specific operation: Sends an HTTP request from the server to the device and executes the interface to display the question on the screen.

[0273] Step 7:

[0274] The user answers the authentication question displayed on the terminal. For example, the user answers "Tanaka."

[0275] Input: The question displayed and the user's answer

[0276] Output: User response data

[0277] What happens: The user enters their answers in text format into the device, and the device collects the data.

[0278] Step 8:

[0279] The device sends the collected user responses to the server, which evaluates the responses and uses a generative AI model to determine whether the user is legitimate.

[0280] Input: User response data

[0281] Output: User authentication evaluation result

[0282] What it does: The server analyzes the user's answers and evaluates them to ensure their validity using a generative AI model.

[0283] Step 9:

[0284] The server notifies the terminal of the evaluation result of the user authentication, and the terminal displays the result to the user. For example, if the authentication is successful, it displays "Authentication successful" to notify the user that suspicion of fraud has been cleared.

[0285] Input: User authentication evaluation result

[0286] Output: Notification message displayed on the terminal

[0287] Specific operation: The server sends the notification result to the device, and the device updates the UI elements to display the result to the user.

[0288] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0289] This invention combines an emotion engine with a system that detects potential fraud based on voice input and authenticates users as needed to achieve more accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[0290] System configuration and operation

[0291] 1. Collecting and converting voice input

[0292] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[0293] 2. Keyword detection and context analysis

[0294] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[0295] 3. Emotion Recognition and Analysis

[0296] The server has an emotion engine for recognizing the user's emotions from the voice input. The emotion engine analyzes the tone, speed, accent, etc. of the voice to determine whether the user is in an emotional state such as anxiety, tension, excitement, etc.

[0297] 4. Generating and Displaying User Authentication Questions

[0298] If the server determines that fraud is likely or that the user is in a particular emotional state (e.g., anxiety or nervousness), it generates a user authentication challenge, which is specific to the individual user, such as information about the user's relatives. The challenge is then sent to the user's device and displayed.

[0299] 5. Collecting and Evaluating User Responses

[0300] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[0301] Specific examples

[0302] The specific flow when the user says "transfer money" is shown below.

[0303] 1. User: Say "transfer money" into the terminal.

[0304] 2. Device: Records audio and sends the data to the server.

[0305] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[0306] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[0307] 5. Server: Uses an emotion engine to recognize whether the user is anxious or nervous.

[0308] 6. Server: Based on a comprehensive assessment of the likelihood of fraud and the emotional state, the server determines that the risk is high.

[0309] 7. Server: Generate the authentication question "What is your mother's maiden name?"

[0310] 8. Server: Sends the question to the user's terminal and displays it.

[0311] 9. User: Answers "Tanaka."

[0312] 10. Terminal: Send this response to the server.

[0313] 11. Server: Evaluate whether "Tanaka" is legitimate.

[0314] 12. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[0315] In this way, the system analyzes user comments and emotions through multiple steps to prevent fraud.

[0316] The processing flow will be explained below.

[0317] Step 1:

[0318] The user makes a statement such as "transfer money" through voice input.

[0319] Step 2:

[0320] The terminal records the user's voice in real time and saves it as voice data.

[0321] Step 3:

[0322] The terminal transmits the recorded voice data to the server.

[0323] Step 4:

[0324] The server converts the received voice data into text using a voice recognition module.

[0325] Step 5:

[0326] The server detects specific keywords (e.g., "transfer money") in the converted text.

[0327] Step 6:

[0328] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[0329] Step 7:

[0330] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[0331] Step 8:

[0332] The server analyzes the voice data using an emotion engine to recognize the user's emotional state (e.g., anxiety, tension, excitement).

[0333] Step 9:

[0334] The server evaluates whether the user is in a particular emotional state based on the output of the emotion engine.

[0335] Step 10:

[0336] The server comprehensively assesses the fraud risk and the emotional state and determines that there is a high possibility of fraudulent activity.

[0337] Step 11:

[0338] The server generates a user authentication question, such as "What is your mother's maiden name?"

[0339] Step 12:

[0340] The server sends the generated authentication challenge to the user's terminal.

[0341] Step 13:

[0342] The terminal displays the authentication challenge received from the server to the user.

[0343] Step 14:

[0344] The user answers the displayed authentication question (e.g., answers "Tanaka").

[0345] Step 15:

[0346] The terminal transmits the user's answer to the server.

[0347] Step 16:

[0348] The server evaluates the received user answers and verifies their validity.

[0349] Step 17:

[0350] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[0351] Example 2

[0352] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0353] Conventional systems have limitations in the accuracy of fraud detection based on voice input, resulting in the possibility of false positives and false negatives. Furthermore, because they do not evaluate the user's emotional state, they are unable to respond to situations where an emotional approach would be expected to improve accuracy. This creates the risk of overlooking sophisticated fraud.

[0354] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0355] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating questions for user authentication when there is a high possibility of fraud, means for displaying the generated questions, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, means for recognizing the user's emotions from the voice and assessing the user's emotional state, and means for comprehensively assessing the possibility of fraud and the emotional state. This enables highly accurate detection of fraud and user authentication by analyzing both the voice input and the emotional state.

[0356] "Voice input" refers to the voice that the user speaks into the terminal, and this voice is the voice data collected by the system.

[0357] "Means of collection" refers to the function of capturing voice data using a microphone or sensor built into the device.

[0358] "Means for converting to text" refers to the function of converting collected voice data into text data using voice recognition technology.

[0359] "Specific Keywords" refers to specific words or phrases that the system considers important for fraud detection and contextual analysis.

[0360] "Means for analyzing context" refers to the function of using natural language processing technology to understand the sentences before and after specific keywords and evaluate their meaning and relevance.

[0361] "Means for assessing the likelihood of fraud" refers to a function that assesses the risk level of fraud as a number or score based on the results of contextual analysis.

[0362] "User authentication questions" refer to questions generated to verify a user's identity when fraud is deemed likely.

[0363] "Means for generating" refers to algorithms or models for automatically creating user authentication questions, in particular generative AI models.

[0364] The "display means" refers to a function for displaying the generated question on the user's terminal screen.

[0365] "Means for collecting answers" refers to the functionality for incorporating answers provided by users into the system.

[0366] "Means for evaluation" refers to the function of analyzing collected responses and determining whether they are from legitimate users.

[0367] "Means for recognizing emotions" refers to technology for analyzing a user's emotional state from voice data, evaluating the tone, speed, accent, etc. of the voice.

[0368] The "means for evaluating emotional state" refers to a function for evaluating the user's psychological state as a numerical value or a category based on the recognized emotional data.

[0369] "Means for making a comprehensive judgment" refers to the function of taking into consideration both the assessment of the possibility of fraud and the assessment of emotional state to make a final judgment.

[0370] This invention is a system that detects possible fraudulent activity based on voice input and authenticates users as necessary. Furthermore, this system is combined with an emotion engine to achieve highly accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[0371] System Configuration and Operation

[0372] Hardware and Software Overview

[0373] 1. Collecting voice input

[0374] The user uses the built-in microphone of the terminal to input voice commands such as "transfer money."

[0375] The terminal transmits the collected voice data to the server.

[0376] The server converts the audio data into text using the Google Cloud Speech-to-Text API.

[0377] 2. Keyword detection and context analysis

[0378] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data.

[0379] If keywords are detected, the server performs a contextual analysis using natural language processing techniques (e.g., spaCy) to assess the likelihood of fraud.

[0380] The likelihood of fraud is scored, for example, rated as 80%.

[0381] 3. Emotion recognition

[0382] The server uses an emotion recognition engine such as DeepAffects to recognize the user's emotions from their voice.

[0383] It analyzes the tone, speed and accent of the voice to determine whether the user is anxious or nervous.

[0384] 4. Generating and displaying user authentication questions

[0385] If fraudulent activity is deemed likely or the user is in a particular emotional state, the server uses a generative AI model (e.g., GPT-3®) to generate user authentication questions.

[0386] Sample question: "What is your mother's maiden name?"

[0387] The server sends the generated question to the user's terminal, where it is displayed.

[0388] 5. Collecting and Evaluating User Answers

[0389] The user answers the displayed question, for example, by replying "Tanaka."

[0390] The terminal sends this response to the server.

[0391] The server analyzes the response and checks the database to see if the user is legitimate.

[0392] The authentication result is notified to the terminal, and if the suspicion of fraud is cleared, a message indicating successful authentication is displayed on the terminal.

[0393] Specific examples

[0394] When the user says "transfer money", the system behaves as follows:

[0395] 1. User: Say "transfer money" into the terminal.

[0396] 2. Device: Records audio using the built-in microphone and sends the audio data to the server.

[0397] 3. Server: Convert the audio data into text using the Google Cloud Speech-to-Text API.

[0398] 4. Server: Detects the keyword "transfer money" from the converted text.

[0399] 5. Server: Analyze the context with spaCy and score the likelihood of fraud at 80%.

[0400] 6. Server: Analyze the user's emotional state from their voice using DeepAffects. Recognize emotions such as anxiety and tension.

[0401] 7. Server: Use the GPT-3 model to generate the authentication question, "What is your mother's maiden name?"

[0402] 8. Server: Sends the question to the user's device and displays it on the device.

[0403] 9. User: Answers "Tanaka."

[0404] 10. Terminal: Send this response to the server.

[0405] 11. Server: Evaluates the answers against a database.

[0406] 12. Server: The authentication result is notified to the terminal, and if there is no suspicion of fraud, a message indicating successful authentication is displayed.

[0407] The system enables highly accurate fraud detection and user authentication through voice and emotion analysis.

[0408] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0409] Step 1:

[0410] The user uses the built-in microphone of the device to input voice commands such as "transfer money." This becomes the input data (voice data).

[0411] Step 2:

[0412] The device records the voice data collected by the built-in microphone. Then, this voice data is sent to the server, and the sent voice data becomes the input data for the server.

[0413] Step 3:

[0414] The server converts the received voice data into text data using a voice recognition module (e.g., Google Cloud Speech-to-Text API). The input is voice data, and the output is text data.

[0415] Step 4:

[0416] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data. The input is the text data, and the output is the keyword detection results and their location information.

[0417] Step 5:

[0418] If a keyword is detected, the server uses natural language processing (NLP) technology (e.g., spaCy) to analyze the context and evaluate the likelihood of fraud. The input is text data and keyword detection results, and the output is a score indicating the likelihood of fraud.

[0419] Step 6:

[0420] The server uses emotion recognition technology (e.g., DeepAffects) to analyze the user's emotions from the voice data. The input is the voice data, and the output is the evaluation result of the emotional state. Specifically, it analyzes voice characteristics such as tone, speed, and accent.

[0421] Step 7:

[0422] The server comprehensively assesses the fraud likelihood assessment and emotional state assessment, and if it determines that the risk is high, it uses a generative AI model (e.g., GPT-3) to generate questions for user authentication. The input is the fraud likelihood score and the emotional state assessment result, and the output is questions for user authentication. Questions that are generated include, "What is your mother's maiden name?"

[0423] Step 8:

[0424] The server sends the generated question to the user's terminal, and the terminal displays the question on the screen. The input is the question for user authentication, and the output is the question displayed on the terminal.

[0425] Step 9:

[0426] The user answers the question displayed on the terminal. For example, the answer "Tanaka" is used. This becomes the input data.

[0427] Step 10:

[0428] The terminal collects the user's answers and sends them to the server. The input is the user's answers, and the output is the answer data sent to the server.

[0429] Step 11:

[0430] The server analyzes the received response data and compares it with a database to determine whether the user is a legitimate user. The input is the user's response data, and the output is the authentication result.

[0431] Step 12:

[0432] The server notifies the user's terminal of the authentication result. The terminal displays this authentication result to the user. The input is the authentication result, and the output is the authentication result displayed on the terminal. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the terminal.

[0433] (Application example 2)

[0434] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0435] Conventional fraud detection systems rely solely on simple keyword detection using voice recognition, resulting in frequent inaccuracies. Furthermore, they fail to consider the user's emotional state, resulting in insufficient fraud risk assessment. The present invention aims to address these issues and achieve more accurate fraud detection and user authentication.

[0436] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0437] In this invention, the server includes means for collecting voice input, means for analyzing emotional state from the voice data, and means for generating user authentication challenges when fraud is likely, thereby taking into account the user's emotional state and providing real-time warnings of potential fraud, as well as providing immediate additional user authentication if necessary.

[0438] A "means for collecting voice input" is a device or software capable of recording user-uttered speech on the device and transmitting it to a server for analysis.

[0439] A "means for converting collected voice input to text" is a process or device that uses voice recognition technology to convert voice data into written data.

[0440] The "means for detecting specific keywords in text" refers to a device or software that has the function of identifying and extracting predetermined keywords from the converted text data.

[0441] "Means for analyzing the context surrounding detected keywords and assessing the likelihood of fraud" refers to a process or device that performs a detailed analysis of the context surrounding keywords and assesses the risk of fraud based on that content.

[0442] A "means for generating user authentication questions when fraud is likely" is a process or device that automatically generates additional user verification questions when a high risk of fraud is determined.

[0443] The "means for displaying the generated question" is a device or software that provides an interface for displaying the generated authentication question on the user's terminal.

[0444] The "means for collecting user answers" is a process or device that collects the answers given by the user to the authentication questions and transmits them to the server.

[0445] The "means for evaluating collected answers and determining whether a user is legitimate" is a process or device that analyzes collected answers and evaluates their legitimacy.

[0446] The "means for analyzing emotional state from voice data" is a process or device that analyzes the tone, rate, accent, etc. of the voice to determine the emotional state of the user.

[0447] A "means for alerting a user to the possibility of fraud in real time while the user is on a call" is a device or software that has the function of assessing the risk of fraud in real time while the user is on a call and immediately notifying the user of the results.

[0448] This invention is a system that detects possible fraud based on voice input and authenticates users as needed, and by combining it with an emotion engine, achieves more accurate fraud detection and user authentication. The invention is implemented by a combination of specific hardware and software.

[0449] Hardware and software used

[0450] 1. Smartphones: Collect voice input.

[0451] 2. Speech recognition software (e.g. speech_recognition package): converts voice data into text.

[0452] 3. Emotion recognition API (e.g. emotion-detect API): Analyzes the user's emotional state.

[0453] 4. Natural language processing engine: Detects specific keywords from text data and analyzes the context.

[0454] Processing flow

[0455] 1. A means of converting collected voice input into text

[0456] Voice data sent by a user into a smartphone is converted into text data using voice recognition software.

[0457] 2. How to find specific keywords

[0458] Detect specified keywords (e.g., "transfer," "remittance," "money") from the converted text data.

[0459] 3. A means of analyzing context and assessing potential fraud

[0460] The context surrounding the detected keywords is analyzed using a natural language processing engine to score the likelihood of fraud.

[0461] 4. A means of analyzing emotional states

[0462] The collected voice data is sent to an emotion recognition API to analyze the user's emotional state (e.g., anxiety, tension).

[0463] 5. Generating User Authentication Challenges When Fraud is Highly Likely

[0464] If fraud is deemed likely, the system generates a user authentication challenge based on the user's personal information, taking into account the user's emotional state.

[0465] 6. A way to display the generated questions

[0466] The generated question is displayed on the smartphone, prompting the user to answer.

[0467] 7. How to collect user responses

[0468] The user's answers are collected and sent to the server.

[0469] 8. A means of evaluating collected responses and determining whether the user is legitimate

[0470] The server evaluates the user's answer and determines whether it is valid.

[0471] Specific examples

[0472] For example, if a user says "transfer money," the voice input is recorded and converted to text using speech recognition software. Corresponding keywords are then detected, and context and sentiment analysis is performed. If the server determines that the user is anxious or nervous, along with the possibility of fraud, it generates an additional authentication question, "What is your mother's maiden name?", which is displayed on the user's smartphone. If the user enters the answer "Tanaka," the answer is sent to the server and evaluated for validity.

[0473] Prompt Sentence Examples

[0474] "Analyze the voice of someone saying they're transferring money, analyze the emotion, and if there's a possibility of fraud, generate additional authentication questions for the user."

[0475] As such, the present invention is a system that combines voice recognition, natural language processing, and emotion analysis to achieve highly accurate fraud detection and user authentication in real time.

[0476] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0477] Step 1:

[0478] The user collects voice input.

[0479] Input: Voice data spoken by the user into the smartphone.

[0480] What happens: The user says, "Transfer money." The smartphone's microphone records this speech and generates a digital audio file.

[0481] Output: Collected audio data.

[0482] Step 2:

[0483] The device sends the voice data to a server and uses a means to convert the voice input into text.

[0484] Input: Audio data.

[0485] Specific operation: The smartphone sends the recorded voice data to the server, where it is converted into text data using voice recognition software (e.g., Google Speech Recognition API).

[0486] Output: Text data.

[0487] Step 3:

[0488] The server uses a means for detecting specific keywords from the text data.

[0489] Input: Text data.

[0490] Specific operation: The server's natural language processing engine searches and detects keywords such as "transfer," "remittance," and "money" within the text data.

[0491] Output: The detected keywords.

[0492] Step 4:

[0493] The server uses a method to analyze the context of the keywords and assess the likelihood of fraud.

[0494] Input: The detected keyword and its surrounding text context.

[0495] How it works: The natural language processing engine analyzes the context surrounding the keyword and scores the likelihood of fraud based on that content. For example, it evaluates the rationality of the action of "transferring money" from the context.

[0496] Output: A score indicating the likelihood of fraud.

[0497] Step 5:

[0498] The server uses a means for analyzing the emotional state from the voice data.

[0499] Input: Audio data.

[0500] Specific operation: The server sends the voice data to an emotion recognition API (e.g., emotion-detect API) and analyzes the emotional state (e.g., anxiety, tension).

[0501] Output: User's emotional state data.

[0502] Step 6:

[0503] The server uses a means to assess the likelihood of fraud and emotional state and generate a user authentication challenge if fraud is likely.

[0504] Input: Deception score, user emotional state data.

[0505] What it does: The server evaluates the fraud score and emotional state data together, and generates user authentication questions (e.g., "What is your mother's maiden name?") if the risk is deemed high.

[0506] Output: The generated authentication questions.

[0507] Step 7:

[0508] The terminal uses a means for displaying the generated question.

[0509] Input: The generated authentication question.

[0510] Specific operation: The server generates a question, which is sent to the smartphone and displayed on the smartphone screen.

[0511] Output: A smartphone screen showing the authentication questions.

[0512] Step 8:

[0513] The user provides answers to questions using means collected by the terminal.

[0514] Input: The user's answer.

[0515] Specific operation: The user enters an answer (e.g., "Tanaka") to the displayed question, and the smartphone collects the answer and sends it to the server.

[0516] Output: Collected response data.

[0517] Step 9:

[0518] The server evaluates the collected answers and uses a means to determine if the user is legitimate.

[0519] Input: Collected response data.

[0520] Specific operation: The server checks the user's answer against a pre-registered database to evaluate its validity. If the answer is valid, it sends a message to the smartphone indicating successful authentication.

[0521] Output: Data indicating the success or failure of the authentication, and a message based on the result.

[0522] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0523] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (registered trademark) (Internet search engine).<URL: https: / / openai.com / blog / chatgpt> ), Gemini (registered trademark) (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0524] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0525] [Second embodiment]

[0526] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0527] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0528] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0529] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0530] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0531] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0532] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0533] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0534] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0535] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0536] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0537] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal."

[0538] This invention relates to a system that detects potential fraudulent activity based on voice input and, if necessary, authenticates the user. This system is realized by combining speech recognition technology and natural language processing (NLP).

[0539] System configuration and operation

[0540] 1. Collecting and converting voice input

[0541] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[0542] 2. Keyword detection and context analysis

[0543] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[0544] 3. Generating and Displaying Authentication Questions

[0545] If fraud is deemed likely, the server generates a challenge, which is tailored to the individual user, such as information about the user's relatives, and is sent to the user's device for display.

[0546] 4. Collecting and Evaluating User Responses

[0547] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[0548] Specific examples

[0549] The specific flow when the user says "transfer money" is shown below.

[0550] 1. User: Say "transfer money" into the terminal.

[0551] 2. Device: Records audio and sends the data to the server.

[0552] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[0553] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[0554] 5. Server: Generate the authentication question "What is your mother's maiden name?"

[0555] 6. Server: Sends the question to the user's terminal and displays it.

[0556] 7. User: Answers "Tanaka."

[0557] 8. Terminal: Send this response to the server.

[0558] 9. Server: Evaluate whether "Tanaka" is legitimate.

[0559] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[0560] In this way, the system analyzes user statements through multiple steps to prevent fraud.

[0561] The processing flow will be explained below.

[0562] Step 1:

[0563] The user makes a statement such as "transfer money" through voice input.

[0564] Step 2:

[0565] The terminal records the user's voice in real time and saves it as voice data.

[0566] Step 3:

[0567] The terminal transmits the recorded voice data to the server.

[0568] Step 4:

[0569] The server converts the received voice data into text using a voice recognition module.

[0570] Step 5:

[0571] The server detects specific keywords (e.g., "transfer money") in the converted text.

[0572] Step 6:

[0573] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[0574] Step 7:

[0575] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[0576] Step 8:

[0577] The server generates a user authentication question, such as "What is your mother's maiden name?"

[0578] Step 9:

[0579] The server sends the generated authentication challenge to the user's terminal.

[0580] Step 10:

[0581] The terminal displays the authentication challenge received from the server to the user.

[0582] Step 11:

[0583] The user answers the displayed authentication question (e.g., answers "Tanaka").

[0584] Step 12:

[0585] The terminal transmits the user's answer to the server.

[0586] Step 13:

[0587] The server evaluates the received user answers and verifies their validity.

[0588] Step 14:

[0589] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[0590] Example 1

[0591] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0592] While user interfaces using voice input have become widespread in recent years, fraudulent acts by malicious third parties have also increased. It is particularly important to prevent fraudulent acts when transactions are made using voice input, such as when exchanging personal information. However, conventional voice recognition systems have not been able to adequately detect fraudulent acts and authenticate users, and a highly reliable authentication system is needed.

[0593] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0594] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating user authentication questions if the possibility of fraud is high, means for generating user authentication questions using a generative AI model, means for transmitting the generated questions to a user terminal and displaying them, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, and means for notifying the user terminal of the authentication result. This enables fraud detection and prevention through advanced voice input analysis and authentication processes.

[0595] "Voice input" refers to recording what the user says as voice data on the terminal.

[0596] "Voice data" refers to the digital signal obtained from voice input, and is data that is sent to a server for analysis.

[0597] "Text data" refers to character string information converted from voice data using voice recognition technology.

[0598] A "keyword" is a key word or phrase extracted from voice input or text data based on specific conditions.

[0599] "Contextual analysis" is the process of evaluating the context within text data and analyzing its overall meaning and relevance.

[0600] "Possibility of fraud" refers to a score or probability that evaluates the likelihood that the content of voice input or text data constitutes fraud.

[0601] An "authentication challenge" is a question generated to verify the authenticity of a user.

[0602] A "generative AI model" is an artificial intelligence that uses machine learning techniques to generate new questions and text.

[0603] A "user terminal" is a device that allows a user to input voice and receive authentication questions, and includes a smartphone, computer, etc.

[0604] "User authentication" is the process of determining whether a user is legitimate based on collected information (e.g., answers to authentication questions).

[0605] The "authentication result" is the result of user authentication, and is information indicating whether the authentication was successful or unsuccessful.

[0606] MODE FOR CARRYING OUT THE INVENTION

[0607] This invention relates to a system that detects possible fraudulent activity based on voice input and authenticates the user as necessary. This system is realized by combining voice recognition technology and natural language processing (NLP). Specific embodiments for carrying out the invention are described below.

[0608] System configuration

[0609] This system mainly consists of a server, a terminal, and a user. Each component functions as follows:

[0610] Collecting and sending voice input

[0611] The user speaks to the terminal, saying something like "transfer money." The terminal records the voice using a built-in microphone and generates voice data. This terminal can be a smartphone, tablet, or PC. The recorded voice data is sent to the server using HTTPS as the communication protocol.

[0612] Audio data conversion

[0613] The server converts the received voice data into text data using a speech recognition module such as Google Cloud Speech-to-Text API. For example, if someone says "transfer money," the voice data will be converted into text data that reads "transfer money."

[0614] Keyword detection and context analysis

[0615] The server receives the text data and uses the natural language processing library spaCy to detect specific keywords within the text. If a keyword is detected, the server analyzes the context and assesses the likelihood of fraud by assigning a score ranging from 0% to 100%.

[0616] Generate and submit authentication questions

[0617] If it determines that there is a high possibility of fraud, the server uses a generative AI model (e.g., GPT-4) to generate a question for user authentication. For example, a question such as "What is your mother's maiden name?" may be generated. This question includes information about the user's close relationships. The generated question is sent to the user's device and displayed on the device screen.

[0618] Collecting and evaluating user responses

[0619] The user enters an answer to the displayed authentication question. The device collects this answer and sends it to the server. The server receives the user's answer and compares it with a pre-configured correct answer database to evaluate whether it is valid. For example, if the user answers "Tanaka," it verifies whether the answer is valid.

[0620] Notification of authentication results

[0621] The server determines whether the authentication was successful based on the user's answers and notifies the user terminal of the result. If the authentication was successful, a message to that effect is displayed on the terminal screen.

[0622] Specific examples

[0623] A specific example in which the user says "transfer money" will be shown below.

[0624] 1. User: Say "transfer money" into the terminal.

[0625] 2. Device: Records audio and sends the data to the server.

[0626] 3. Server: Convert the audio data into text using Google Cloud Speech-to-Text. Detect the phrase "transfer money."

[0627] 4. Server: Uses spaCy to analyze the context and scores the likelihood of fraud at 80%.

[0628] 5. Server: Use GPT-4 to generate the authentication question "What is your mother's maiden name?"

[0629] 6. Server: Sends the question to the user's terminal and displays it.

[0630] 7. User: Answers "Tanaka."

[0631] 8. Terminal: Send this response to the server.

[0632] 9. Server: Evaluate whether "Tanaka" is legitimate.

[0633] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[0634] In this way, the system prevents fraud through advanced voice input analysis and authentication processes.

[0635] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0636] Step 1:

[0637] User: The user speaks into the terminal, saying "transfer money."

[0638] Specific action: Record audio using a microphone.

[0639] Input: User's voice

[0640] Output: Recorded audio data

[0641] Step 2:

[0642] Device: The device collects the recorded voice data and sends it to the server.

[0643] Specific operation: The recorded audio data is sent to the server via HTTPS.

[0644] Input: Recorded audio data

[0645] Output: Audio data sent to the server

[0646] Step 3:

[0647] Server: The server converts the received voice data into text data using a speech recognition module (e.g., Google Cloud Speech-to-Text).

[0648] Specific operation: Sends audio data to the API and receives it as text data.

[0649] Input: Audio data

[0650] Output: Text data

[0651] Step 4:

[0652] Server: The server uses a natural language processing library (e.g., spaCy) to detect whether specific keywords (e.g., "transfer money") are contained in the text data.

[0653] Specific operation: Matches text data with a keyword list and extracts matching keywords.

[0654] Input: Text data

[0655] Output: Text data containing keywords

[0656] Step 5:

[0657] Server: If a keyword is detected, the context is analyzed and a score is assigned for potential fraud.

[0658] What it does: Runs a contextual analysis algorithm and scores the likelihood of fraud on a scale of 0% to 100%.

[0659] Input: Text data containing keywords

[0660] Output: A score indicating the likelihood of fraud.

[0661] Step 6:

[0662] Server: If fraud is deemed likely, a generative AI model (e.g., GPT-4) is used to generate user authentication questions.

[0663] Specific operation: The AI ​​model is given a prompt, "Generate a question using information about the user's close relationships," and the generated question is received.

[0664] Input: Fraud likelihood score, prompt statement

[0665] Output: Generated authentication questions

[0666] Step 7:

[0667] Server: Sends the generated authentication question to the user's terminal and instructs it to be displayed.

[0668] Specific operation: The text data of the question is sent to the terminal via HTTPS.

[0669] Input: Generated authentication question

[0670] Output: Authentication question displayed on the user's terminal

[0671] Step 8:

[0672] User: The user enters the answer to the displayed authentication question. For example, the user answers "Tanaka."

[0673] Specific actions: Enter your answer in the input field on the device and press the send button.

[0674] Input: Answer to authentication question

[0675] Output: Answers typed into the terminal

[0676] Step 9:

[0677] Terminal: The terminal collects the user's answers and sends them to the server.

[0678] Specific operation: The entered answer is sent to the server via HTTPS.

[0679] Input: Answer typed into the terminal

[0680] Output: The answer sent to the server

[0681] Step 10:

[0682] Server: Compares the received user answers with a pre-configured database of correct answers and evaluates whether the user is legitimate.

[0683] Specific behavior: Compares the answer data with the correct answer database to see if it matches.

[0684] Input: User answers, correct answer database

[0685] Output: Evaluation result (valid or not)

[0686] Step 11:

[0687] Server: Generates the authentication result and notifies the user terminal.

[0688] Specific operation: Creates a message indicating whether authentication was successful or failed and sends it to the terminal via HTTPS communication.

[0689] Input: Evaluation result

[0690] Output: Authentication result sent to the terminal

[0691] Step 12:

[0692] Terminal: Displays the authentication result received from the server to the user.

[0693] Specific behavior: A message indicating authentication success or failure is displayed on the screen.

[0694] Input: Authentication result

[0695] Output: Authentication result displayed to the user

[0696] (Application example 1)

[0697] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0698] Current voice input systems have difficulty detecting fraud in real time and quickly authenticating users. Especially in electronic payments, where transactions are frequent and fast, there is a lack of means to prevent fraud immediately. Furthermore, fraud detection systems based on voice input often rely too much on specific keywords, making it difficult to accurately analyze the context, leading to false positives. Therefore, a system that can detect fraud and authenticate users with higher accuracy is needed.

[0699] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0700] In this invention, the server includes means for collecting voice input in real time, means for converting the voice input into text, means for detecting specific keywords in the text, means for analyzing the context and assessing the likelihood of fraud, means for generating additional authentication challenges if the likelihood of fraud is high, and means for displaying the generated challenges on the user terminal, thereby enabling instant fraud detection using voice and rapid user authentication.

[0701] "Voice input" refers to spoken content or instructions from a user through a voice collection device such as a microphone.

[0702] "Means for converting to text" refers to speech recognition technology or software for converting voice input into character string data.

[0703] "Means for detecting specific keywords" refers to the ability to identify and extract predefined important words and phrases within the converted text.

[0704] "Contextual fraud risk assessment" refers to algorithms that analyze the context of text and the use of language to score or assess the risk of fraud.

[0705] The "means for generating additional authentication questions" refers to the ability to create new questions to verify the user's authenticity if it is determined that there is a high possibility of fraud.

[0706] "Means for displaying on the user terminal" refers to the technology or interface for displaying the generated authentication question on the screen of the device used by the user.

[0707] "Means for recognizing voice in real time" refers to a function that instantly detects and recognizes the user's speech and then processes it.

[0708] "Means for obtaining analysis results on the server" refers to the function of obtaining the results of speech recognition and text analysis from a remote server and sharing and using them throughout the system.

[0709] A "generative AI model" is a machine learning model that uses artificial intelligence to analyze and process input data and generate appropriate output.

[0710] A "prompt" is a document containing instructions or questions to be input into a generative AI model, providing criteria and conditions for data analysis.

[0711] DETAILED DESCRIPTION OF THE INVENTION The following describes an embodiment of the present invention.

[0712] The server has a means for collecting voice input, and collects the voice spoken by the user into the terminal. The collected voice is converted into text using voice recognition technology. An example of the voice recognition technology used here is Google's voice recognition API. By using this API, the collected voice data can be converted into text in real time.

[0713] The converted text is sent to a server where it is processed to detect specific keywords. Natural Language Processing (NLP) techniques are recommended to detect specific keywords in the text. Once a specific keyword is detected, the server analyzes the context and assesses its likelihood of fraud. This assessment is performed using a generative AI model, which analyzes the prompt sentence as input.

[0714] If the server determines that fraud is likely, it generates additional questions for user authentication. These questions may contain information about the user's relatives, for example, and are important for verifying the user's authenticity. The generated questions are sent to the user's terminal and displayed on its screen.

[0715] The user answers the displayed questions, and the answers are sent to the server via the device. The server evaluates the user's answers and determines whether the user is legitimate. This authentication uses a generative AI model to instantly evaluate the validity of the answers.

[0716] Hardware or software used

[0717] Hardware: Smartphone, microphone

[0718] software:

[0719] Speech Recognition API (Google)

[0720] Natural Language Processing (NLP)

[0721] Generative AI Models

[0722] Libraries for handling HTTP requests (requests)

[0723] Realization of processing steps

[0724] For example, if the user says "I would like to purchase this item," the following steps are performed:

[0725] 1. Audio collection and conversion:

[0726] The user speaks into their smartphone, saying, "I would like to purchase this product." The device collects this speech and converts it into text using a speech recognition API.

[0727] 2. Text transmission and analysis:

[0728] The text is sent to a server, which detects specific keywords (such as "purchase") and analyzes the context. If it is determined to be fraudulent, the corresponding prompt is fed into a generative AI model.

[0729] 3. Generate and display authentication questions:

[0730] If the generative AI model scores a high probability of fraud, it generates additional authentication questions and displays them on the user's device, such as "What is your mother's maiden name?"

[0731] 4. User answers and their ratings:

[0732] If the user answers "Tanaka," the device sends the answer to the server, which evaluates the answer using a generative AI model to confirm its validity.

[0733] Prompt Sentence Examples

[0734] An example of a prompt used in this system is as follows:

[0735] "Please enter the following information into the system:

[0736] Say: "Buy this item."

[0737] Contextual analysis results: 80% probability of fraud.

[0738] Verification Question: What is your mother's maiden name?

[0739] User Answer: Tanaka.

[0740] Please evaluate the user for legitimacy and notify us of the results."

[0741] This embodiment allows for instant fraud detection and highly accurate user authentication based on voice input.

[0742] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0743] Step 1:

[0744] A user speaks into a device (smartphone) to input information. For example, the user might say, "I would like to purchase this product."

[0745] Input: User's voice data

[0746] Output: Collected audio data

[0747] Step 2:

[0748] The device sends the collected voice data in real time to a voice recognition API (for example, Google's voice recognition API), which converts the voice data into text data.

[0749] Input: Collected audio data

[0750] Output: Converted text data

[0751] Specific operation: Sends audio data to the API and receives text data from the API.

[0752] Step 3:

[0753] The device transmits the converted text data to a server, which performs text analysis to detect specific keywords in the text data, for example, the keyword "purchase."

[0754] Input: Text data

[0755] Output: Keyword detection results

[0756] What it does: Runs a text analysis algorithm to extract specific keywords.

[0757] Step 4:

[0758] The server analyzes the context of the detected keywords and utilizes a generative AI model to assess the likelihood of fraud. The prompt sentence is provided as input to the generative AI model, which scores the probability of fraud.

[0759] Input: Text data and detected keywords

[0760] Output: Fraud likelihood score

[0761] Specific operation: Contextual analysis is performed using NLP technology, prompt sentences are input into the generative AI model, and a score is calculated.

[0762] Step 5:

[0763] If the fraud score is high, the server generates additional questions to authenticate the user, for example, "What is your mother's maiden name?"

[0764] Input: Fraud likelihood score

[0765] Output: Generated authentication questions

[0766] What it does: Selects and generates appropriate questions from a generative AI model or a predefined question set.

[0767] Step 6:

[0768] The generated user authentication question is sent from the server to the terminal and displayed on the terminal screen.

[0769] Input: Generated authentication question

[0770] Output: Question displayed on terminal

[0771] Specific operation: Sends an HTTP request from the server to the device and executes the interface to display the question on the screen.

[0772] Step 7:

[0773] The user answers the authentication question displayed on the terminal. For example, the user answers "Tanaka."

[0774] Input: The question displayed and the user's answer

[0775] Output: User response data

[0776] What happens: The user enters their answers in text format into the device, and the device collects the data.

[0777] Step 8:

[0778] The device sends the collected user responses to the server, which evaluates the responses and uses a generative AI model to determine whether the user is legitimate.

[0779] Input: User response data

[0780] Output: User authentication evaluation result

[0781] What it does: The server analyzes the user's answers and evaluates them to ensure their validity using a generative AI model.

[0782] Step 9:

[0783] The server notifies the terminal of the evaluation result of the user authentication, and the terminal displays the result to the user. For example, if the authentication is successful, it displays "Authentication successful" to notify the user that suspicion of fraud has been cleared.

[0784] Input: User authentication evaluation result

[0785] Output: Notification message displayed on the terminal

[0786] Specific operation: The server sends the notification result to the device, and the device updates the UI elements to display the result to the user.

[0787] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0788] This invention combines an emotion engine with a system that detects potential fraud based on voice input and authenticates users as needed to achieve more accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[0789] System configuration and operation

[0790] 1. Collecting and converting voice input

[0791] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[0792] 2. Keyword detection and context analysis

[0793] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[0794] 3. Emotion Recognition and Analysis

[0795] The server has an emotion engine for recognizing the user's emotions from the voice input. The emotion engine analyzes the tone, speed, accent, etc. of the voice to determine whether the user is in an emotional state such as anxiety, tension, excitement, etc.

[0796] 4. Generating and Displaying User Authentication Questions

[0797] If the server determines that fraud is likely or that the user is in a particular emotional state (e.g., anxiety or nervousness), it generates a user authentication challenge, which is specific to the individual user, such as information about the user's relatives. The challenge is then sent to the user's device and displayed.

[0798] 5. Collecting and Evaluating User Responses

[0799] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[0800] Specific examples

[0801] The specific flow when the user says "transfer money" is shown below.

[0802] 1. User: Say "transfer money" into the terminal.

[0803] 2. Device: Records audio and sends the data to the server.

[0804] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[0805] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[0806] 5. Server: Uses an emotion engine to recognize whether the user is anxious or nervous.

[0807] 6. Server: Based on a comprehensive assessment of the likelihood of fraud and the emotional state, the server determines that the risk is high.

[0808] 7. Server: Generate the authentication question "What is your mother's maiden name?"

[0809] 8. Server: Sends the question to the user's terminal and displays it.

[0810] 9. User: Answers "Tanaka."

[0811] 10. Terminal: Send this response to the server.

[0812] 11. Server: Evaluate whether "Tanaka" is legitimate.

[0813] 12. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[0814] In this way, the system analyzes user comments and emotions through multiple steps to prevent fraud.

[0815] The processing flow will be explained below.

[0816] Step 1:

[0817] The user makes a statement such as "transfer money" through voice input.

[0818] Step 2:

[0819] The terminal records the user's voice in real time and saves it as voice data.

[0820] Step 3:

[0821] The terminal transmits the recorded voice data to the server.

[0822] Step 4:

[0823] The server converts the received voice data into text using a voice recognition module.

[0824] Step 5:

[0825] The server detects specific keywords (e.g., "transfer money") in the converted text.

[0826] Step 6:

[0827] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[0828] Step 7:

[0829] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[0830] Step 8:

[0831] The server analyzes the voice data using an emotion engine to recognize the user's emotional state (e.g., anxiety, tension, excitement).

[0832] Step 9:

[0833] The server evaluates whether the user is in a particular emotional state based on the output of the emotion engine.

[0834] Step 10:

[0835] The server comprehensively assesses the fraud risk and the emotional state and determines that there is a high possibility of fraudulent activity.

[0836] Step 11:

[0837] The server generates a user authentication question, such as "What is your mother's maiden name?"

[0838] Step 12:

[0839] The server sends the generated authentication challenge to the user's terminal.

[0840] Step 13:

[0841] The terminal displays the authentication challenge received from the server to the user.

[0842] Step 14:

[0843] The user answers the displayed authentication question (e.g., answers "Tanaka").

[0844] Step 15:

[0845] The terminal transmits the user's answer to the server.

[0846] Step 16:

[0847] The server evaluates the received user answers and verifies their validity.

[0848] Step 17:

[0849] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[0850] Example 2

[0851] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0852] Conventional systems have limitations in the accuracy of fraud detection based on voice input, resulting in the possibility of false positives and false negatives. Furthermore, because they do not evaluate the user's emotional state, they are unable to respond to situations where an emotional approach would be expected to improve accuracy. This creates the risk of overlooking sophisticated fraud.

[0853] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0854] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating questions for user authentication when there is a high possibility of fraud, means for displaying the generated questions, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, means for recognizing the user's emotions from the voice and assessing the user's emotional state, and means for comprehensively assessing the possibility of fraud and the emotional state. This enables highly accurate detection of fraud and user authentication by analyzing both the voice input and the emotional state.

[0855] "Voice input" refers to the voice that the user speaks into the terminal, and this voice is the voice data collected by the system.

[0856] "Means of collection" refers to the function of capturing voice data using a microphone or sensor built into the device.

[0857] "Means for converting to text" refers to the function of converting collected voice data into text data using voice recognition technology.

[0858] "Specific Keywords" refers to specific words or phrases that the system considers important for fraud detection and contextual analysis.

[0859] "Means for analyzing context" refers to the function of using natural language processing technology to understand the sentences before and after specific keywords and evaluate their meaning and relevance.

[0860] "Means for assessing the likelihood of fraud" refers to a function that assesses the risk level of fraud as a number or score based on the results of contextual analysis.

[0861] "User authentication questions" refer to questions generated to verify a user's identity when fraud is deemed likely.

[0862] "Means for generating" refers to algorithms or models for automatically creating user authentication questions, in particular generative AI models.

[0863] The "display means" refers to a function for displaying the generated question on the user's terminal screen.

[0864] "Means for collecting answers" refers to the functionality for incorporating answers provided by users into the system.

[0865] "Means for evaluation" refers to the function of analyzing collected responses and determining whether they are from legitimate users.

[0866] "Means for recognizing emotions" refers to technology for analyzing a user's emotional state from voice data, evaluating the tone, speed, accent, etc. of the voice.

[0867] The "means for evaluating emotional state" refers to a function for evaluating the user's psychological state as a numerical value or a category based on the recognized emotional data.

[0868] "Means for making a comprehensive judgment" refers to the function of taking into consideration both the assessment of the possibility of fraud and the assessment of emotional state to make a final judgment.

[0869] This invention is a system that detects possible fraudulent activity based on voice input and authenticates users as necessary. Furthermore, this system is combined with an emotion engine to achieve highly accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[0870] System Configuration and Operation

[0871] Hardware and Software Overview

[0872] 1. Collecting voice input

[0873] The user uses the built-in microphone of the terminal to input voice commands such as "transfer money."

[0874] The terminal transmits the collected voice data to the server.

[0875] The server converts the audio data into text using the Google Cloud Speech-to-Text API.

[0876] 2. Keyword detection and context analysis

[0877] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data.

[0878] If keywords are detected, the server performs a contextual analysis using natural language processing techniques (e.g., spaCy) to assess the likelihood of fraud.

[0879] The likelihood of fraud is scored, for example, rated as 80%.

[0880] 3. Emotion recognition

[0881] The server uses an emotion recognition engine such as DeepAffects to recognize the user's emotions from their voice.

[0882] It analyzes the tone, speed and accent of the voice to determine whether the user is anxious or nervous.

[0883] 4. Generating and displaying user authentication questions

[0884] If fraud is deemed likely or the user is in a particular emotional state, the server uses a generative AI model (e.g., GPT-3) to generate user authentication questions.

[0885] Sample question: "What is your mother's maiden name?"

[0886] The server sends the generated question to the user's terminal, where it is displayed.

[0887] 5. Collecting and Evaluating User Answers

[0888] The user answers the displayed question, for example, by replying "Tanaka."

[0889] The terminal sends this response to the server.

[0890] The server analyzes the response and checks the database to see if the user is legitimate.

[0891] The authentication result is notified to the terminal, and if the suspicion of fraud is cleared, a message indicating successful authentication is displayed on the terminal.

[0892] Specific examples

[0893] When the user says "transfer money", the system behaves as follows:

[0894] 1. User: Say "transfer money" into the terminal.

[0895] 2. Device: Records audio using the built-in microphone and sends the audio data to the server.

[0896] 3. Server: Convert the audio data into text using the Google Cloud Speech-to-Text API.

[0897] 4. Server: Detects the keyword "transfer money" from the converted text.

[0898] 5. Server: Analyze the context with spaCy and score the likelihood of fraud at 80%.

[0899] 6. Server: Analyze the user's emotional state from their voice using DeepAffects. Recognize emotions such as anxiety and tension.

[0900] 7. Server: Use the GPT-3 model to generate the authentication question, "What is your mother's maiden name?"

[0901] 8. Server: Sends the question to the user's device and displays it on the device.

[0902] 9. User: Answers "Tanaka."

[0903] 10. Terminal: Send this response to the server.

[0904] 11. Server: Evaluates the answers against a database.

[0905] 12. Server: The authentication result is notified to the terminal, and if there is no suspicion of fraud, a message indicating successful authentication is displayed.

[0906] The system enables highly accurate fraud detection and user authentication through voice and emotion analysis.

[0907] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0908] Step 1:

[0909] The user uses the built-in microphone of the device to input voice commands such as "transfer money." This becomes the input data (voice data).

[0910] Step 2:

[0911] The device records the voice data collected by the built-in microphone. Then, this voice data is sent to the server, and the sent voice data becomes the input data for the server.

[0912] Step 3:

[0913] The server converts the received voice data into text data using a voice recognition module (e.g., Google Cloud Speech-to-Text API). The input is voice data, and the output is text data.

[0914] Step 4:

[0915] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data. The input is the text data, and the output is the keyword detection results and their location information.

[0916] Step 5:

[0917] If a keyword is detected, the server uses natural language processing (NLP) technology (e.g., spaCy) to analyze the context and evaluate the likelihood of fraud. The input is text data and keyword detection results, and the output is a score indicating the likelihood of fraud.

[0918] Step 6:

[0919] The server uses emotion recognition technology (e.g., DeepAffects) to analyze the user's emotions from the voice data. The input is the voice data, and the output is the evaluation result of the emotional state. Specifically, it analyzes voice characteristics such as tone, speed, and accent.

[0920] Step 7:

[0921] The server comprehensively assesses the fraud likelihood assessment and emotional state assessment, and if it determines that the risk is high, it uses a generative AI model (e.g., GPT-3) to generate questions for user authentication. The input is the fraud likelihood score and the emotional state assessment result, and the output is questions for user authentication. Questions that are generated include, "What is your mother's maiden name?"

[0922] Step 8:

[0923] The server sends the generated question to the user's terminal, and the terminal displays the question on the screen. The input is the question for user authentication, and the output is the question displayed on the terminal.

[0924] Step 9:

[0925] The user answers the question displayed on the terminal. For example, the answer "Tanaka" is used. This becomes the input data.

[0926] Step 10:

[0927] The terminal collects the user's answers and sends them to the server. The input is the user's answers, and the output is the answer data sent to the server.

[0928] Step 11:

[0929] The server analyzes the received response data and compares it with a database to determine whether the user is a legitimate user. The input is the user's response data, and the output is the authentication result.

[0930] Step 12:

[0931] The server notifies the user's terminal of the authentication result. The terminal displays this authentication result to the user. The input is the authentication result, and the output is the authentication result displayed on the terminal. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the terminal.

[0932] (Application example 2)

[0933] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0934] Conventional fraud detection systems rely solely on simple keyword detection using voice recognition, resulting in frequent inaccuracies. Furthermore, they fail to consider the user's emotional state, resulting in insufficient fraud risk assessment. The present invention aims to address these issues and achieve more accurate fraud detection and user authentication.

[0935] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[0936] In this invention, the server includes means for collecting voice input, means for analyzing emotional state from the voice data, and means for generating user authentication challenges when fraud is likely, thereby taking into account the user's emotional state and providing real-time warnings of potential fraud, as well as providing immediate additional user authentication if necessary.

[0937] A "means for collecting voice input" is a device or software capable of recording user-uttered speech on the device and transmitting it to a server for analysis.

[0938] A "means for converting collected voice input to text" is a process or device that uses voice recognition technology to convert voice data into written data.

[0939] The "means for detecting specific keywords in text" refers to a device or software that has the function of identifying and extracting predetermined keywords from the converted text data.

[0940] "Means for analyzing the context surrounding detected keywords and assessing the likelihood of fraud" refers to a process or device that performs a detailed analysis of the context surrounding keywords and assesses the risk of fraud based on that content.

[0941] A "means for generating user authentication questions when fraud is likely" is a process or device that automatically generates additional user verification questions when a high risk of fraud is determined.

[0942] The "means for displaying the generated question" is a device or software that provides an interface for displaying the generated authentication question on the user's terminal.

[0943] The "means for collecting user answers" is a process or device that collects the answers given by the user to the authentication questions and transmits them to the server.

[0944] The "means for evaluating collected answers and determining whether a user is legitimate" is a process or device that analyzes collected answers and evaluates their legitimacy.

[0945] The "means for analyzing emotional state from voice data" is a process or device that analyzes the tone, rate, accent, etc. of the voice to determine the emotional state of the user.

[0946] A "means for alerting a user to the possibility of fraud in real time while the user is on a call" is a device or software that has the function of assessing the risk of fraud in real time while the user is on a call and immediately notifying the user of the results.

[0947] This invention is a system that detects possible fraud based on voice input and authenticates users as needed, and by combining it with an emotion engine, achieves more accurate fraud detection and user authentication. The invention is implemented by a combination of specific hardware and software.

[0948] Hardware and software used

[0949] 1. Smartphones: Collect voice input.

[0950] 2. Speech recognition software (e.g. speech_recognition package): converts voice data into text.

[0951] 3. Emotion recognition API (e.g. emotion-detect API): Analyzes the user's emotional state.

[0952] 4. Natural language processing engine: Detects specific keywords from text data and analyzes the context.

[0953] Processing flow

[0954] 1. A means of converting collected voice input into text

[0955] Voice data sent by a user into a smartphone is converted into text data using voice recognition software.

[0956] 2. How to find specific keywords

[0957] Detect specified keywords (e.g., "transfer," "remittance," "money") from the converted text data.

[0958] 3. A means of analyzing context and assessing potential fraud

[0959] The context surrounding the detected keywords is analyzed using a natural language processing engine to score the likelihood of fraud.

[0960] 4. A means of analyzing emotional states

[0961] The collected voice data is sent to an emotion recognition API to analyze the user's emotional state (e.g., anxiety, tension).

[0962] 5. Generating User Authentication Challenges When Fraud is Highly Likely

[0963] If fraud is deemed likely, the system generates a user authentication challenge based on the user's personal information, taking into account the user's emotional state.

[0964] 6. A way to display the generated questions

[0965] The generated question is displayed on the smartphone, prompting the user to answer.

[0966] 7. How to collect user responses

[0967] The user's answers are collected and sent to the server.

[0968] 8. A means of evaluating collected responses and determining whether the user is legitimate

[0969] The server evaluates the user's answer and determines whether it is valid.

[0970] Specific examples

[0971] For example, if a user says "transfer money," the voice input is recorded and converted to text using speech recognition software. Corresponding keywords are then detected, and context and sentiment analysis is performed. If the server determines that the user is anxious or nervous, along with the possibility of fraud, it generates an additional authentication question, "What is your mother's maiden name?", which is displayed on the user's smartphone. If the user enters the answer "Tanaka," the answer is sent to the server and evaluated for validity.

[0972] Prompt Sentence Examples

[0973] "Analyze the voice of someone saying they're transferring money, analyze the emotion, and if there's a possibility of fraud, generate additional authentication questions for the user."

[0974] As such, the present invention is a system that combines voice recognition, natural language processing, and emotion analysis to achieve highly accurate fraud detection and user authentication in real time.

[0975] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0976] Step 1:

[0977] The user collects voice input.

[0978] Input: Voice data spoken by the user into the smartphone.

[0979] What happens: The user says, "Transfer money." The smartphone's microphone records this speech and generates a digital audio file.

[0980] Output: Collected audio data.

[0981] Step 2:

[0982] The device sends the voice data to a server and uses a means to convert the voice input into text.

[0983] Input: Audio data.

[0984] Specific operation: The smartphone sends the recorded voice data to the server, where it is converted into text data using voice recognition software (e.g., Google Speech Recognition API).

[0985] Output: Text data.

[0986] Step 3:

[0987] The server uses a means for detecting specific keywords from the text data.

[0988] Input: Text data.

[0989] Specific operation: The server's natural language processing engine searches and detects keywords such as "transfer," "remittance," and "money" within the text data.

[0990] Output: The detected keywords.

[0991] Step 4:

[0992] The server uses a method to analyze the context of the keywords and assess the likelihood of fraud.

[0993] Input: The detected keyword and its surrounding text context.

[0994] How it works: The natural language processing engine analyzes the context surrounding the keyword and scores the likelihood of fraud based on that content. For example, it evaluates the rationality of the action of "transferring money" from the context.

[0995] Output: A score indicating the likelihood of fraud.

[0996] Step 5:

[0997] The server uses a means for analyzing the emotional state from the voice data.

[0998] Input: Audio data.

[0999] Specific operation: The server sends the voice data to an emotion recognition API (e.g., emotion-detect API) and analyzes the emotional state (e.g., anxiety, tension).

[1000] Output: User's emotional state data.

[1001] Step 6:

[1002] The server uses a means to assess the likelihood of fraud and emotional state and generate a user authentication challenge if fraud is likely.

[1003] Input: Deception score, user emotional state data.

[1004] What it does: The server evaluates the fraud score and emotional state data together, and generates user authentication questions (e.g., "What is your mother's maiden name?") if the risk is deemed high.

[1005] Output: The generated authentication questions.

[1006] Step 7:

[1007] The terminal uses a means for displaying the generated question.

[1008] Input: The generated authentication question.

[1009] Specific operation: The server generates a question, which is sent to the smartphone and displayed on the smartphone screen.

[1010] Output: A smartphone screen showing the authentication questions.

[1011] Step 8:

[1012] The user provides answers to questions using means collected by the terminal.

[1013] Input: The user's answer.

[1014] Specific operation: The user enters an answer (e.g., "Tanaka") to the displayed question, and the smartphone collects the answer and sends it to the server.

[1015] Output: Collected response data.

[1016] Step 9:

[1017] The server evaluates the collected answers and uses a means to determine if the user is legitimate.

[1018] Input: Collected response data.

[1019] Specific operation: The server checks the user's answer against a pre-registered database to evaluate its validity. If the answer is valid, it sends a message to the smartphone indicating successful authentication.

[1020] Output: Data indicating the success or failure of the authentication, and a message based on the result.

[1021] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1022] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1023] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[1024] [Third embodiment]

[1025] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[1026] 5, the data processing system 310 includes the data processing device 12 and a headset type terminal 314. An example of the data processing device 12 is a server.

[1027] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1028] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[1029] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1030] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1031] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1032] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1033] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1034] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1035] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1036] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the headset type terminal 314 will be referred to as the "terminal."

[1037] This invention relates to a system that detects potential fraudulent activity based on voice input and, if necessary, authenticates the user. This system is realized by combining speech recognition technology and natural language processing (NLP).

[1038] System configuration and operation

[1039] 1. Collecting and converting voice input

[1040] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[1041] 2. Keyword detection and context analysis

[1042] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[1043] 3. Generating and Displaying Authentication Questions

[1044] If fraud is deemed likely, the server generates a challenge, which is tailored to the individual user, such as information about the user's relatives, and is sent to the user's device for display.

[1045] 4. Collecting and Evaluating User Responses

[1046] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[1047] Specific examples

[1048] The specific flow when the user says "transfer money" is shown below.

[1049] 1. User: Say "transfer money" into the terminal.

[1050] 2. Device: Records audio and sends the data to the server.

[1051] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[1052] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[1053] 5. Server: Generate the authentication question "What is your mother's maiden name?"

[1054] 6. Server: Sends the question to the user's terminal and displays it.

[1055] 7. User: Answers "Tanaka."

[1056] 8. Terminal: Send this response to the server.

[1057] 9. Server: Evaluate whether "Tanaka" is legitimate.

[1058] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[1059] In this way, the system analyzes user statements through multiple steps to prevent fraud.

[1060] The processing flow will be explained below.

[1061] Step 1:

[1062] The user makes a statement such as "transfer money" through voice input.

[1063] Step 2:

[1064] The terminal records the user's voice in real time and saves it as voice data.

[1065] Step 3:

[1066] The terminal transmits the recorded voice data to the server.

[1067] Step 4:

[1068] The server converts the received voice data into text using a voice recognition module.

[1069] Step 5:

[1070] The server detects specific keywords (e.g., "transfer money") in the converted text.

[1071] Step 6:

[1072] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[1073] Step 7:

[1074] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[1075] Step 8:

[1076] The server generates a user authentication question, such as "What is your mother's maiden name?"

[1077] Step 9:

[1078] The server sends the generated authentication challenge to the user's terminal.

[1079] Step 10:

[1080] The terminal displays the authentication challenge received from the server to the user.

[1081] Step 11:

[1082] The user answers the displayed authentication question (e.g., answers "Tanaka").

[1083] Step 12:

[1084] The terminal transmits the user's answer to the server.

[1085] Step 13:

[1086] The server evaluates the received user answers and verifies their validity.

[1087] Step 14:

[1088] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[1089] Example 1

[1090] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1091] While user interfaces using voice input have become widespread in recent years, fraudulent acts by malicious third parties have also increased. It is particularly important to prevent fraudulent acts when transactions are made using voice input, such as when exchanging personal information. However, conventional voice recognition systems have not been able to adequately detect fraudulent acts and authenticate users, and a highly reliable authentication system is needed.

[1092] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1093] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating user authentication questions if the possibility of fraud is high, means for generating user authentication questions using a generative AI model, means for transmitting the generated questions to a user terminal and displaying them, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, and means for notifying the user terminal of the authentication result. This enables fraud detection and prevention through advanced voice input analysis and authentication processes.

[1094] "Voice input" refers to recording what the user says as voice data on the terminal.

[1095] "Voice data" refers to the digital signal obtained from voice input, and is data that is sent to a server for analysis.

[1096] "Text data" refers to character string information converted from voice data using voice recognition technology.

[1097] A "keyword" is a key word or phrase extracted from voice input or text data based on specific conditions.

[1098] "Contextual analysis" is the process of evaluating the context within text data and analyzing its overall meaning and relevance.

[1099] "Possibility of fraud" refers to a score or probability that evaluates the likelihood that the content of voice input or text data constitutes fraud.

[1100] An "authentication challenge" is a question generated to verify the authenticity of a user.

[1101] A "generative AI model" is an artificial intelligence that uses machine learning techniques to generate new questions and text.

[1102] A "user terminal" is a device that allows a user to input voice and receive authentication questions, and includes a smartphone, computer, etc.

[1103] "User authentication" is the process of determining whether a user is legitimate based on collected information (e.g., answers to authentication questions).

[1104] The "authentication result" is the result of user authentication, and is information indicating whether the authentication was successful or unsuccessful.

[1105] MODE FOR CARRYING OUT THE INVENTION

[1106] This invention relates to a system that detects possible fraudulent activity based on voice input and authenticates the user as necessary. This system is realized by combining voice recognition technology and natural language processing (NLP). Specific embodiments for carrying out the invention are described below.

[1107] System configuration

[1108] This system mainly consists of a server, a terminal, and a user. Each component functions as follows:

[1109] Collecting and sending voice input

[1110] The user speaks to the terminal, saying something like "transfer money." The terminal records the voice using a built-in microphone and generates voice data. This terminal can be a smartphone, tablet, or PC. The recorded voice data is sent to the server using HTTPS as the communication protocol.

[1111] Audio data conversion

[1112] The server converts the received voice data into text data using a speech recognition module such as Google Cloud Speech-to-Text API. For example, if someone says "transfer money," the voice data will be converted into text data that reads "transfer money."

[1113] Keyword detection and context analysis

[1114] The server receives the text data and uses the natural language processing library spaCy to detect specific keywords within the text. If a keyword is detected, the server analyzes the context and assesses the likelihood of fraud by assigning a score ranging from 0% to 100%.

[1115] Generate and submit authentication questions

[1116] If it determines that there is a high possibility of fraud, the server uses a generative AI model (e.g., GPT-4) to generate a question for user authentication. For example, a question such as "What is your mother's maiden name?" may be generated. This question includes information about the user's close relationships. The generated question is sent to the user's device and displayed on the device screen.

[1117] Collecting and evaluating user responses

[1118] The user enters an answer to the displayed authentication question. The device collects this answer and sends it to the server. The server receives the user's answer and compares it with a pre-configured correct answer database to evaluate whether it is valid. For example, if the user answers "Tanaka," it verifies whether the answer is valid.

[1119] Notification of authentication results

[1120] The server determines whether the authentication was successful based on the user's answers and notifies the user terminal of the result. If the authentication was successful, a message to that effect is displayed on the terminal screen.

[1121] Specific examples

[1122] A specific example in which the user says "transfer money" will be shown below.

[1123] 1. User: Say "transfer money" into the terminal.

[1124] 2. Device: Records audio and sends the data to the server.

[1125] 3. Server: Convert the audio data into text using Google Cloud Speech-to-Text. Detect the phrase "transfer money."

[1126] 4. Server: Uses spaCy to analyze the context and scores the likelihood of fraud at 80%.

[1127] 5. Server: Use GPT-4 to generate the authentication question "What is your mother's maiden name?"

[1128] 6. Server: Sends the question to the user's terminal and displays it.

[1129] 7. User: Answers "Tanaka."

[1130] 8. Terminal: Send this response to the server.

[1131] 9. Server: Evaluate whether "Tanaka" is legitimate.

[1132] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[1133] In this way, the system prevents fraud through advanced voice input analysis and authentication processes.

[1134] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1135] Step 1:

[1136] User: The user speaks into the terminal, saying "transfer money."

[1137] Specific action: Record audio using a microphone.

[1138] Input: User's voice

[1139] Output: Recorded audio data

[1140] Step 2:

[1141] Device: The device collects the recorded voice data and sends it to the server.

[1142] Specific operation: The recorded audio data is sent to the server via HTTPS.

[1143] Input: Recorded audio data

[1144] Output: Audio data sent to the server

[1145] Step 3:

[1146] Server: The server converts the received voice data into text data using a speech recognition module (e.g., Google Cloud Speech-to-Text).

[1147] Specific operation: Sends audio data to the API and receives it as text data.

[1148] Input: Audio data

[1149] Output: Text data

[1150] Step 4:

[1151] Server: The server uses a natural language processing library (e.g., spaCy) to detect whether specific keywords (e.g., "transfer money") are contained in the text data.

[1152] Specific operation: Matches text data with a keyword list and extracts matching keywords.

[1153] Input: Text data

[1154] Output: Text data containing keywords

[1155] Step 5:

[1156] Server: If a keyword is detected, the context is analyzed and a score is assigned for potential fraud.

[1157] What it does: Runs a contextual analysis algorithm and scores the likelihood of fraud on a scale of 0% to 100%.

[1158] Input: Text data containing keywords

[1159] Output: A score indicating the likelihood of fraud.

[1160] Step 6:

[1161] Server: If fraud is deemed likely, a generative AI model (e.g., GPT-4) is used to generate user authentication questions.

[1162] Specific operation: The AI ​​model is given a prompt, "Generate a question using information about the user's close relationships," and the generated question is received.

[1163] Input: Fraud likelihood score, prompt statement

[1164] Output: Generated authentication questions

[1165] Step 7:

[1166] Server: Sends the generated authentication question to the user's terminal and instructs it to be displayed.

[1167] Specific operation: The text data of the question is sent to the terminal via HTTPS.

[1168] Input: Generated authentication question

[1169] Output: Authentication question displayed on the user's terminal

[1170] Step 8:

[1171] User: The user enters the answer to the displayed authentication question. For example, the user answers "Tanaka."

[1172] Specific actions: Enter your answer in the input field on the device and press the send button.

[1173] Input: Answer to authentication question

[1174] Output: Answers typed into the terminal

[1175] Step 9:

[1176] Terminal: The terminal collects the user's answers and sends them to the server.

[1177] Specific operation: The entered answer is sent to the server via HTTPS.

[1178] Input: Answer typed into the terminal

[1179] Output: The answer sent to the server

[1180] Step 10:

[1181] Server: Compares the received user answers with a pre-configured database of correct answers and evaluates whether the user is legitimate.

[1182] Specific behavior: Compares the answer data with the correct answer database to see if it matches.

[1183] Input: User answers, correct answer database

[1184] Output: Evaluation result (valid or not)

[1185] Step 11:

[1186] Server: Generates the authentication result and notifies the user terminal.

[1187] Specific operation: Creates a message indicating whether authentication was successful or failed and sends it to the terminal via HTTPS communication.

[1188] Input: Evaluation result

[1189] Output: Authentication result sent to the terminal

[1190] Step 12:

[1191] Terminal: Displays the authentication result received from the server to the user.

[1192] Specific behavior: A message indicating authentication success or failure is displayed on the screen.

[1193] Input: Authentication result

[1194] Output: Authentication result displayed to the user

[1195] (Application example 1)

[1196] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1197] Current voice input systems have difficulty detecting fraud in real time and quickly authenticating users. Especially in electronic payments, where transactions are frequent and fast, there is a lack of means to prevent fraud immediately. Furthermore, fraud detection systems based on voice input often rely too much on specific keywords, making it difficult to accurately analyze the context, leading to false positives. Therefore, a system that can detect fraud and authenticate users with higher accuracy is needed.

[1198] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1199] In this invention, the server includes means for collecting voice input in real time, means for converting the voice input into text, means for detecting specific keywords in the text, means for analyzing the context and assessing the likelihood of fraud, means for generating additional authentication challenges if the likelihood of fraud is high, and means for displaying the generated challenges on the user terminal, thereby enabling instant fraud detection using voice and rapid user authentication.

[1200] "Voice input" refers to spoken content or instructions from a user through a voice collection device such as a microphone.

[1201] "Means for converting to text" refers to speech recognition technology or software for converting voice input into character string data.

[1202] "Means for detecting specific keywords" refers to the ability to identify and extract predefined important words and phrases within the converted text.

[1203] "Contextual fraud risk assessment" refers to algorithms that analyze the context of text and the use of language to score or assess the risk of fraud.

[1204] The "means for generating additional authentication questions" refers to the ability to create new questions to verify the user's authenticity if it is determined that there is a high possibility of fraud.

[1205] "Means for displaying on the user terminal" refers to the technology or interface for displaying the generated authentication question on the screen of the device used by the user.

[1206] "Means for recognizing voice in real time" refers to a function that instantly detects and recognizes the user's speech and then processes it.

[1207] "Means for obtaining analysis results on the server" refers to the function of obtaining the results of speech recognition and text analysis from a remote server and sharing and using them throughout the system.

[1208] A "generative AI model" is a machine learning model that uses artificial intelligence to analyze and process input data and generate appropriate output.

[1209] A "prompt" is a document containing instructions or questions to be input into a generative AI model, providing criteria and conditions for data analysis.

[1210] DETAILED DESCRIPTION OF THE INVENTION The following describes an embodiment of the present invention.

[1211] The server has a means for collecting voice input, and collects the voice spoken by the user into the terminal. The collected voice is converted into text using voice recognition technology. An example of the voice recognition technology used here is Google's voice recognition API. By using this API, the collected voice data can be converted into text in real time.

[1212] The converted text is sent to a server where it is processed to detect specific keywords. Natural Language Processing (NLP) techniques are recommended to detect specific keywords in the text. Once a specific keyword is detected, the server analyzes the context and assesses its likelihood of fraud. This assessment is performed using a generative AI model, which analyzes the prompt sentence as input.

[1213] If the server determines that fraud is likely, it generates additional questions for user authentication. These questions may contain information about the user's relatives, for example, and are important for verifying the user's authenticity. The generated questions are sent to the user's terminal and displayed on its screen.

[1214] The user answers the displayed questions, and the answers are sent to the server via the device. The server evaluates the user's answers and determines whether the user is legitimate. This authentication uses a generative AI model to instantly evaluate the validity of the answers.

[1215] Hardware or software used

[1216] Hardware: Smartphone, microphone

[1217] software:

[1218] Speech Recognition API (Google)

[1219] Natural Language Processing (NLP)

[1220] Generative AI Models

[1221] Libraries for handling HTTP requests (requests)

[1222] Realization of processing steps

[1223] For example, if the user says "I would like to purchase this item," the following steps are performed:

[1224] 1. Audio collection and conversion:

[1225] The user speaks into their smartphone, saying, "I would like to purchase this product." The device collects this speech and converts it into text using a speech recognition API.

[1226] 2. Text transmission and analysis:

[1227] The text is sent to a server, which detects specific keywords (such as "purchase") and analyzes the context. If it is determined to be fraudulent, the corresponding prompt is fed into a generative AI model.

[1228] 3. Generate and display authentication questions:

[1229] If the generative AI model scores a high probability of fraud, it generates additional authentication questions and displays them on the user's device, such as "What is your mother's maiden name?"

[1230] 4. User answers and their ratings:

[1231] If the user answers "Tanaka," the device sends the answer to the server, which evaluates the answer using a generative AI model to confirm its validity.

[1232] Prompt Sentence Examples

[1233] An example of a prompt used in this system is as follows:

[1234] "Please enter the following information into the system:

[1235] Say: "Buy this item."

[1236] Contextual analysis results: 80% probability of fraud.

[1237] Verification Question: What is your mother's maiden name?

[1238] User Answer: Tanaka.

[1239] Please evaluate the user for legitimacy and notify us of the results."

[1240] This embodiment allows for instant fraud detection and highly accurate user authentication based on voice input.

[1241] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1242] Step 1:

[1243] A user speaks into a device (smartphone) to input information. For example, the user might say, "I would like to purchase this product."

[1244] Input: User's voice data

[1245] Output: Collected audio data

[1246] Step 2:

[1247] The device sends the collected voice data in real time to a voice recognition API (for example, Google's voice recognition API), which converts the voice data into text data.

[1248] Input: Collected audio data

[1249] Output: Converted text data

[1250] Specific operation: Sends audio data to the API and receives text data from the API.

[1251] Step 3:

[1252] The device transmits the converted text data to a server, which performs text analysis to detect specific keywords in the text data, for example, the keyword "purchase."

[1253] Input: Text data

[1254] Output: Keyword detection results

[1255] What it does: Runs a text analysis algorithm to extract specific keywords.

[1256] Step 4:

[1257] The server analyzes the context of the detected keywords and utilizes a generative AI model to assess the likelihood of fraud. The prompt sentence is provided as input to the generative AI model, which scores the probability of fraud.

[1258] Input: Text data and detected keywords

[1259] Output: Fraud likelihood score

[1260] Specific operation: Contextual analysis is performed using NLP technology, prompt sentences are input into the generative AI model, and a score is calculated.

[1261] Step 5:

[1262] If the fraud score is high, the server generates additional questions to authenticate the user, for example, "What is your mother's maiden name?"

[1263] Input: Fraud likelihood score

[1264] Output: Generated authentication questions

[1265] What it does: Selects and generates appropriate questions from a generative AI model or a predefined question set.

[1266] Step 6:

[1267] The generated user authentication question is sent from the server to the terminal and displayed on the terminal screen.

[1268] Input: Generated authentication question

[1269] Output: Question displayed on terminal

[1270] Specific operation: Sends an HTTP request from the server to the device and executes the interface to display the question on the screen.

[1271] Step 7:

[1272] The user answers the authentication question displayed on the terminal. For example, the user answers "Tanaka."

[1273] Input: The question displayed and the user's answer

[1274] Output: User response data

[1275] What happens: The user enters their answers in text format into the device, and the device collects the data.

[1276] Step 8:

[1277] The device sends the collected user responses to the server, which evaluates the responses and uses a generative AI model to determine whether the user is legitimate.

[1278] Input: User response data

[1279] Output: User authentication evaluation result

[1280] What it does: The server analyzes the user's answers and evaluates them to ensure their validity using a generative AI model.

[1281] Step 9:

[1282] The server notifies the terminal of the evaluation result of the user authentication, and the terminal displays the result to the user. For example, if the authentication is successful, it displays "Authentication successful" to notify the user that suspicion of fraud has been cleared.

[1283] Input: User authentication evaluation result

[1284] Output: Notification message displayed on the terminal

[1285] Specific operation: The server sends the notification result to the device, and the device updates the UI elements to display the result to the user.

[1286] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1287] This invention combines an emotion engine with a system that detects potential fraud based on voice input and authenticates users as needed to achieve more accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[1288] System configuration and operation

[1289] 1. Collecting and converting voice input

[1290] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[1291] 2. Keyword detection and context analysis

[1292] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[1293] 3. Emotion Recognition and Analysis

[1294] The server has an emotion engine for recognizing the user's emotions from the voice input. The emotion engine analyzes the tone, speed, accent, etc. of the voice to determine whether the user is in an emotional state such as anxiety, tension, excitement, etc.

[1295] 4. Generating and Displaying User Authentication Questions

[1296] If the server determines that fraud is likely or that the user is in a particular emotional state (e.g., anxiety or nervousness), it generates a user authentication challenge, which is specific to the individual user, such as information about the user's relatives. The challenge is then sent to the user's device and displayed.

[1297] 5. Collecting and Evaluating User Responses

[1298] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[1299] Specific examples

[1300] The specific flow when the user says "transfer money" is shown below.

[1301] 1. User: Say "transfer money" into the terminal.

[1302] 2. Device: Records audio and sends the data to the server.

[1303] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[1304] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[1305] 5. Server: Uses an emotion engine to recognize whether the user is anxious or nervous.

[1306] 6. Server: Based on a comprehensive assessment of the likelihood of fraud and the emotional state, the server determines that the risk is high.

[1307] 7. Server: Generate the authentication question "What is your mother's maiden name?"

[1308] 8. Server: Sends the question to the user's terminal and displays it.

[1309] 9. User: Answers "Tanaka."

[1310] 10. Terminal: Send this response to the server.

[1311] 11. Server: Evaluate whether "Tanaka" is legitimate.

[1312] 12. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[1313] In this way, the system analyzes user comments and emotions through multiple steps to prevent fraud.

[1314] The processing flow will be explained below.

[1315] Step 1:

[1316] The user makes a statement such as "transfer money" through voice input.

[1317] Step 2:

[1318] The terminal records the user's voice in real time and saves it as voice data.

[1319] Step 3:

[1320] The terminal transmits the recorded voice data to the server.

[1321] Step 4:

[1322] The server converts the received voice data into text using a voice recognition module.

[1323] Step 5:

[1324] The server detects specific keywords (e.g., "transfer money") in the converted text.

[1325] Step 6:

[1326] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[1327] Step 7:

[1328] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[1329] Step 8:

[1330] The server analyzes the voice data using an emotion engine to recognize the user's emotional state (e.g., anxiety, tension, excitement).

[1331] Step 9:

[1332] The server evaluates whether the user is in a particular emotional state based on the output of the emotion engine.

[1333] Step 10:

[1334] The server comprehensively assesses the fraud risk and the emotional state and determines that there is a high possibility of fraudulent activity.

[1335] Step 11:

[1336] The server generates a user authentication question, such as "What is your mother's maiden name?"

[1337] Step 12:

[1338] The server sends the generated authentication challenge to the user's terminal.

[1339] Step 13:

[1340] The terminal displays the authentication challenge received from the server to the user.

[1341] Step 14:

[1342] The user answers the displayed authentication question (e.g., answers "Tanaka").

[1343] Step 15:

[1344] The terminal transmits the user's answer to the server.

[1345] Step 16:

[1346] The server evaluates the received user answers and verifies their validity.

[1347] Step 17:

[1348] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[1349] Example 2

[1350] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1351] Conventional systems have limitations in the accuracy of fraud detection based on voice input, resulting in the possibility of false positives and false negatives. Furthermore, because they do not evaluate the user's emotional state, they are unable to respond to situations where an emotional approach would be expected to improve accuracy. This creates the risk of overlooking sophisticated fraud.

[1352] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1353] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating questions for user authentication when there is a high possibility of fraud, means for displaying the generated questions, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, means for recognizing the user's emotions from the voice and assessing the user's emotional state, and means for comprehensively assessing the possibility of fraud and the emotional state. This enables highly accurate detection of fraud and user authentication by analyzing both the voice input and the emotional state.

[1354] "Voice input" refers to the voice that the user speaks into the terminal, and this voice is the voice data collected by the system.

[1355] "Means of collection" refers to the function of capturing voice data using a microphone or sensor built into the device.

[1356] "Means for converting to text" refers to the function of converting collected voice data into text data using voice recognition technology.

[1357] "Specific Keywords" refers to specific words or phrases that the system considers important for fraud detection and contextual analysis.

[1358] "Means for analyzing context" refers to the function of using natural language processing technology to understand the sentences before and after specific keywords and evaluate their meaning and relevance.

[1359] "Means for assessing the likelihood of fraud" refers to a function that assesses the risk level of fraud as a number or score based on the results of contextual analysis.

[1360] "User authentication questions" refer to questions generated to verify a user's identity when fraud is deemed likely.

[1361] "Means for generating" refers to algorithms or models for automatically creating user authentication questions, in particular generative AI models.

[1362] The "display means" refers to a function for displaying the generated question on the user's terminal screen.

[1363] "Means for collecting answers" refers to the functionality for incorporating answers provided by users into the system.

[1364] "Means for evaluation" refers to the function of analyzing collected responses and determining whether they are from legitimate users.

[1365] "Means for recognizing emotions" refers to technology for analyzing a user's emotional state from voice data, evaluating the tone, speed, accent, etc. of the voice.

[1366] The "means for evaluating emotional state" refers to a function for evaluating the user's psychological state as a numerical value or a category based on the recognized emotional data.

[1367] "Means for making a comprehensive judgment" refers to the function of taking into consideration both the assessment of the possibility of fraud and the assessment of emotional state to make a final judgment.

[1368] This invention is a system that detects possible fraudulent activity based on voice input and authenticates users as necessary. Furthermore, this system is combined with an emotion engine to achieve highly accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[1369] System Configuration and Operation

[1370] Hardware and Software Overview

[1371] 1. Collecting voice input

[1372] The user uses the built-in microphone of the terminal to input voice commands such as "transfer money."

[1373] The terminal transmits the collected voice data to the server.

[1374] The server converts the audio data into text using the Google Cloud Speech-to-Text API.

[1375] 2. Keyword detection and context analysis

[1376] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data.

[1377] If keywords are detected, the server performs a contextual analysis using natural language processing techniques (e.g., spaCy) to assess the likelihood of fraud.

[1378] The likelihood of fraud is scored, for example, rated as 80%.

[1379] 3. Emotion recognition

[1380] The server uses an emotion recognition engine such as DeepAffects to recognize the user's emotions from their voice.

[1381] It analyzes the tone, speed and accent of the voice to determine whether the user is anxious or nervous.

[1382] 4. Generating and displaying user authentication questions

[1383] If fraud is deemed likely or the user is in a particular emotional state, the server uses a generative AI model (e.g., GPT-3) to generate user authentication questions.

[1384] Sample question: "What is your mother's maiden name?"

[1385] The server sends the generated question to the user's terminal, where it is displayed.

[1386] 5. Collecting and Evaluating User Answers

[1387] The user answers the displayed question, for example, by replying "Tanaka."

[1388] The terminal sends this response to the server.

[1389] The server analyzes the response and checks the database to see if the user is legitimate.

[1390] The authentication result is notified to the terminal, and if the suspicion of fraud is cleared, a message indicating successful authentication is displayed on the terminal.

[1391] Specific examples

[1392] When the user says "transfer money", the system behaves as follows:

[1393] 1. User: Say "transfer money" into the terminal.

[1394] 2. Device: Records audio using the built-in microphone and sends the audio data to the server.

[1395] 3. Server: Convert the audio data into text using the Google Cloud Speech-to-Text API.

[1396] 4. Server: Detects the keyword "transfer money" from the converted text.

[1397] 5. Server: Analyze the context with spaCy and score the likelihood of fraud at 80%.

[1398] 6. Server: Analyze the user's emotional state from their voice using DeepAffects. Recognize emotions such as anxiety and tension.

[1399] 7. Server: Use the GPT-3 model to generate the authentication question, "What is your mother's maiden name?"

[1400] 8. Server: Sends the question to the user's device and displays it on the device.

[1401] 9. User: Answers "Tanaka."

[1402] 10. Terminal: Send this response to the server.

[1403] 11. Server: Evaluates the answers against a database.

[1404] 12. Server: The authentication result is notified to the terminal, and if there is no suspicion of fraud, a message indicating successful authentication is displayed.

[1405] The system enables highly accurate fraud detection and user authentication through voice and emotion analysis.

[1406] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1407] Step 1:

[1408] The user uses the built-in microphone of the device to input voice commands such as "transfer money." This becomes the input data (voice data).

[1409] Step 2:

[1410] The device records the voice data collected by the built-in microphone. Then, this voice data is sent to the server, and the sent voice data becomes the input data for the server.

[1411] Step 3:

[1412] The server converts the received voice data into text data using a voice recognition module (e.g., Google Cloud Speech-to-Text API). The input is voice data, and the output is text data.

[1413] Step 4:

[1414] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data. The input is the text data, and the output is the keyword detection results and their location information.

[1415] Step 5:

[1416] If a keyword is detected, the server uses natural language processing (NLP) technology (e.g., spaCy) to analyze the context and evaluate the likelihood of fraud. The input is text data and keyword detection results, and the output is a score indicating the likelihood of fraud.

[1417] Step 6:

[1418] The server uses emotion recognition technology (e.g., DeepAffects) to analyze the user's emotions from the voice data. The input is the voice data, and the output is the evaluation result of the emotional state. Specifically, it analyzes voice characteristics such as tone, speed, and accent.

[1419] Step 7:

[1420] The server comprehensively assesses the fraud likelihood assessment and emotional state assessment, and if it determines that the risk is high, it uses a generative AI model (e.g., GPT-3) to generate questions for user authentication. The input is the fraud likelihood score and the emotional state assessment result, and the output is questions for user authentication. Questions that are generated include, "What is your mother's maiden name?"

[1421] Step 8:

[1422] The server sends the generated question to the user's terminal, and the terminal displays the question on the screen. The input is the question for user authentication, and the output is the question displayed on the terminal.

[1423] Step 9:

[1424] The user answers the question displayed on the terminal. For example, the answer "Tanaka" is used. This becomes the input data.

[1425] Step 10:

[1426] The terminal collects the user's answers and sends them to the server. The input is the user's answers, and the output is the answer data sent to the server.

[1427] Step 11:

[1428] The server analyzes the received response data and compares it with a database to determine whether the user is a legitimate user. The input is the user's response data, and the output is the authentication result.

[1429] Step 12:

[1430] The server notifies the user's terminal of the authentication result. The terminal displays this authentication result to the user. The input is the authentication result, and the output is the authentication result displayed on the terminal. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the terminal.

[1431] (Application example 2)

[1432] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1433] Conventional fraud detection systems rely solely on simple keyword detection using voice recognition, resulting in frequent inaccuracies. Furthermore, they fail to consider the user's emotional state, resulting in insufficient fraud risk assessment. The present invention aims to address these issues and achieve more accurate fraud detection and user authentication.

[1434] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1435] In this invention, the server includes means for collecting voice input, means for analyzing emotional state from the voice data, and means for generating user authentication challenges when fraud is likely, thereby taking into account the user's emotional state and providing real-time warnings of potential fraud, as well as providing immediate additional user authentication if necessary.

[1436] A "means for collecting voice input" is a device or software capable of recording user-uttered speech on the device and transmitting it to a server for analysis.

[1437] A "means for converting collected voice input to text" is a process or device that uses voice recognition technology to convert voice data into written data.

[1438] The "means for detecting specific keywords in text" refers to a device or software that has the function of identifying and extracting predetermined keywords from the converted text data.

[1439] "Means for analyzing the context surrounding detected keywords and assessing the likelihood of fraud" refers to a process or device that performs a detailed analysis of the context surrounding keywords and assesses the risk of fraud based on that content.

[1440] A "means for generating user authentication questions when fraud is likely" is a process or device that automatically generates additional user verification questions when a high risk of fraud is determined.

[1441] The "means for displaying the generated question" is a device or software that provides an interface for displaying the generated authentication question on the user's terminal.

[1442] The "means for collecting user answers" is a process or device that collects the answers given by the user to the authentication questions and transmits them to the server.

[1443] The "means for evaluating collected answers and determining whether a user is legitimate" is a process or device that analyzes collected answers and evaluates their legitimacy.

[1444] The "means for analyzing emotional state from voice data" is a process or device that analyzes the tone, rate, accent, etc. of the voice to determine the emotional state of the user.

[1445] A "means for alerting a user to the possibility of fraud in real time while the user is on a call" is a device or software that has the function of assessing the risk of fraud in real time while the user is on a call and immediately notifying the user of the results.

[1446] This invention is a system that detects possible fraud based on voice input and authenticates users as needed, and by combining it with an emotion engine, achieves more accurate fraud detection and user authentication. The invention is implemented by a combination of specific hardware and software.

[1447] Hardware and software used

[1448] 1. Smartphones: Collect voice input.

[1449] 2. Speech recognition software (e.g. speech_recognition package): converts voice data into text.

[1450] 3. Emotion recognition API (e.g. emotion-detect API): Analyzes the user's emotional state.

[1451] 4. Natural language processing engine: Detects specific keywords from text data and analyzes the context.

[1452] Processing flow

[1453] 1. A means of converting collected voice input into text

[1454] Voice data sent by a user into a smartphone is converted into text data using voice recognition software.

[1455] 2. How to find specific keywords

[1456] Detect specified keywords (e.g., "transfer," "remittance," "money") from the converted text data.

[1457] 3. A means of analyzing context and assessing potential fraud

[1458] The context surrounding the detected keywords is analyzed using a natural language processing engine to score the likelihood of fraud.

[1459] 4. A means of analyzing emotional states

[1460] The collected voice data is sent to an emotion recognition API to analyze the user's emotional state (e.g., anxiety, tension).

[1461] 5. Generating User Authentication Challenges When Fraud is Highly Likely

[1462] If fraud is deemed likely, the system generates a user authentication challenge based on the user's personal information, taking into account the user's emotional state.

[1463] 6. A way to display the generated questions

[1464] The generated question is displayed on the smartphone, prompting the user to answer.

[1465] 7. How to collect user responses

[1466] The user's answers are collected and sent to the server.

[1467] 8. A means of evaluating collected responses and determining whether the user is legitimate

[1468] The server evaluates the user's answer and determines whether it is valid.

[1469] Specific examples

[1470] For example, if a user says "transfer money," the voice input is recorded and converted to text using speech recognition software. Corresponding keywords are then detected, and context and sentiment analysis is performed. If the server determines that the user is anxious or nervous, along with the possibility of fraud, it generates an additional authentication question, "What is your mother's maiden name?", which is displayed on the user's smartphone. If the user enters the answer "Tanaka," the answer is sent to the server and evaluated for validity.

[1471] Prompt Sentence Examples

[1472] "Analyze the voice of someone saying they're transferring money, analyze the emotion, and if there's a possibility of fraud, generate additional authentication questions for the user."

[1473] As such, the present invention is a system that combines voice recognition, natural language processing, and emotion analysis to achieve highly accurate fraud detection and user authentication in real time.

[1474] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1475] Step 1:

[1476] The user collects voice input.

[1477] Input: Voice data spoken by the user into the smartphone.

[1478] What happens: The user says, "Transfer money." The smartphone's microphone records this speech and generates a digital audio file.

[1479] Output: Collected audio data.

[1480] Step 2:

[1481] The device sends the voice data to a server and uses a means to convert the voice input into text.

[1482] Input: Audio data.

[1483] Specific operation: The smartphone sends the recorded voice data to the server, where it is converted into text data using voice recognition software (e.g., Google Speech Recognition API).

[1484] Output: Text data.

[1485] Step 3:

[1486] The server uses a means for detecting specific keywords from the text data.

[1487] Input: Text data.

[1488] Specific operation: The server's natural language processing engine searches and detects keywords such as "transfer," "remittance," and "money" within the text data.

[1489] Output: The detected keywords.

[1490] Step 4:

[1491] The server uses a method to analyze the context of the keywords and assess the likelihood of fraud.

[1492] Input: The detected keyword and its surrounding text context.

[1493] How it works: The natural language processing engine analyzes the context surrounding the keyword and scores the likelihood of fraud based on that content. For example, it evaluates the rationality of the action of "transferring money" from the context.

[1494] Output: A score indicating the likelihood of fraud.

[1495] Step 5:

[1496] The server uses a means for analyzing the emotional state from the voice data.

[1497] Input: Audio data.

[1498] Specific operation: The server sends the voice data to an emotion recognition API (e.g., emotion-detect API) and analyzes the emotional state (e.g., anxiety, tension).

[1499] Output: User's emotional state data.

[1500] Step 6:

[1501] The server uses a means to assess the likelihood of fraud and emotional state and generate a user authentication challenge if fraud is likely.

[1502] Input: Deception score, user emotional state data.

[1503] What it does: The server evaluates the fraud score and emotional state data together, and generates user authentication questions (e.g., "What is your mother's maiden name?") if the risk is deemed high.

[1504] Output: The generated authentication questions.

[1505] Step 7:

[1506] The terminal uses a means for displaying the generated question.

[1507] Input: The generated authentication question.

[1508] Specific operation: The server generates a question, which is sent to the smartphone and displayed on the smartphone screen.

[1509] Output: A smartphone screen showing the authentication questions.

[1510] Step 8:

[1511] The user provides answers to questions using means collected by the terminal.

[1512] Input: The user's answer.

[1513] Specific operation: The user enters an answer (e.g., "Tanaka") to the displayed question, and the smartphone collects the answer and sends it to the server.

[1514] Output: Collected response data.

[1515] Step 9:

[1516] The server evaluates the collected answers and uses a means to determine if the user is legitimate.

[1517] Input: Collected response data.

[1518] Specific operation: The server checks the user's answer against a pre-registered database to evaluate its validity. If the answer is valid, it sends a message to the smartphone indicating successful authentication.

[1519] Output: Data indicating the success or failure of the authentication, and a message based on the result.

[1520] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1521] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1522] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1523] [Fourth embodiment]

[1524] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1525] 7, a data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1526] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1527] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the control target 443 are also connected to the bus 52.

[1528] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1529] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1530] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1531] The control object 443 includes a display device, LEDs in the eyes, and motors for driving the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1532] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1533] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1534] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1535] In the robot 414, the processor 46 performs the reception output process. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1536] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1537] This invention relates to a system that detects potential fraudulent activity based on voice input and, if necessary, authenticates the user. This system is realized by combining speech recognition technology and natural language processing (NLP).

[1538] System configuration and operation

[1539] 1. Collecting and converting voice input

[1540] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[1541] 2. Keyword detection and context analysis

[1542] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[1543] 3. Generating and Displaying Authentication Questions

[1544] If fraud is deemed likely, the server generates a challenge, which is tailored to the individual user, such as information about the user's relatives, and is sent to the user's device for display.

[1545] 4. Collecting and Evaluating User Responses

[1546] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[1547] Specific examples

[1548] The specific flow when the user says "transfer money" is shown below.

[1549] 1. User: Say "transfer money" into the terminal.

[1550] 2. Device: Records audio and sends the data to the server.

[1551] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[1552] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[1553] 5. Server: Generate the authentication question "What is your mother's maiden name?"

[1554] 6. Server: Sends the question to the user's terminal and displays it.

[1555] 7. User: Answers "Tanaka."

[1556] 8. Terminal: Send this response to the server.

[1557] 9. Server: Evaluate whether "Tanaka" is legitimate.

[1558] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[1559] In this way, the system analyzes user statements through multiple steps to prevent fraud.

[1560] The processing flow will be explained below.

[1561] Step 1:

[1562] The user makes a statement such as "transfer money" through voice input.

[1563] Step 2:

[1564] The terminal records the user's voice in real time and saves it as voice data.

[1565] Step 3:

[1566] The terminal transmits the recorded voice data to the server.

[1567] Step 4:

[1568] The server converts the received voice data into text using a voice recognition module.

[1569] Step 5:

[1570] The server detects specific keywords (e.g., "transfer money") in the converted text.

[1571] Step 6:

[1572] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[1573] Step 7:

[1574] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[1575] Step 8:

[1576] The server generates a user authentication question, such as "What is your mother's maiden name?"

[1577] Step 9:

[1578] The server sends the generated authentication challenge to the user's terminal.

[1579] Step 10:

[1580] The terminal displays the authentication challenge received from the server to the user.

[1581] Step 11:

[1582] The user answers the displayed authentication question (e.g., answers "Tanaka").

[1583] Step 12:

[1584] The terminal transmits the user's answer to the server.

[1585] Step 13:

[1586] The server evaluates the received user answers and verifies their validity.

[1587] Step 14:

[1588] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[1589] Example 1

[1590] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1591] While user interfaces using voice input have become widespread in recent years, fraudulent acts by malicious third parties have also increased. It is particularly important to prevent fraudulent acts when transactions are made using voice input, such as when exchanging personal information. However, conventional voice recognition systems have not been able to adequately detect fraudulent acts and authenticate users, and a highly reliable authentication system is needed.

[1592] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1593] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating user authentication questions if the possibility of fraud is high, means for generating user authentication questions using a generative AI model, means for transmitting the generated questions to a user terminal and displaying them, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, and means for notifying the user terminal of the authentication result. This enables fraud detection and prevention through advanced voice input analysis and authentication processes.

[1594] "Voice input" refers to recording what the user says as voice data on the terminal.

[1595] "Voice data" refers to the digital signal obtained from voice input, and is data that is sent to a server for analysis.

[1596] "Text data" refers to character string information converted from voice data using voice recognition technology.

[1597] A "keyword" is a key word or phrase extracted from voice input or text data based on specific conditions.

[1598] "Contextual analysis" is the process of evaluating the context within text data and analyzing its overall meaning and relevance.

[1599] "Possibility of fraud" refers to a score or probability that evaluates the likelihood that the content of voice input or text data constitutes fraud.

[1600] An "authentication challenge" is a question generated to verify the authenticity of a user.

[1601] A "generative AI model" is an artificial intelligence that uses machine learning techniques to generate new questions and text.

[1602] A "user terminal" is a device that allows a user to input voice and receive authentication questions, and includes a smartphone, computer, etc.

[1603] "User authentication" is the process of determining whether a user is legitimate based on collected information (e.g., answers to authentication questions).

[1604] The "authentication result" is the result of user authentication, and is information indicating whether the authentication was successful or unsuccessful.

[1605] MODE FOR CARRYING OUT THE INVENTION

[1606] This invention relates to a system that detects possible fraudulent activity based on voice input and authenticates the user as necessary. This system is realized by combining voice recognition technology and natural language processing (NLP). Specific embodiments for carrying out the invention are described below.

[1607] System configuration

[1608] This system mainly consists of a server, a terminal, and a user. Each component functions as follows:

[1609] Collecting and sending voice input

[1610] The user speaks to the terminal, saying something like "transfer money." The terminal records the voice using a built-in microphone and generates voice data. This terminal can be a smartphone, tablet, or PC. The recorded voice data is sent to the server using HTTPS as the communication protocol.

[1611] Audio data conversion

[1612] The server converts the received voice data into text data using a speech recognition module such as Google Cloud Speech-to-Text API. For example, if someone says "transfer money," the voice data will be converted into text data that reads "transfer money."

[1613] Keyword detection and context analysis

[1614] The server receives the text data and uses the natural language processing library spaCy to detect specific keywords within the text. If a keyword is detected, the server analyzes the context and assesses the likelihood of fraud by assigning a score ranging from 0% to 100%.

[1615] Generate and submit authentication questions

[1616] If it determines that there is a high possibility of fraud, the server uses a generative AI model (e.g., GPT-4) to generate a question for user authentication. For example, a question such as "What is your mother's maiden name?" may be generated. This question includes information about the user's close relationships. The generated question is sent to the user's device and displayed on the device screen.

[1617] Collecting and evaluating user responses

[1618] The user enters an answer to the displayed authentication question. The device collects this answer and sends it to the server. The server receives the user's answer and compares it with a pre-configured correct answer database to evaluate whether it is valid. For example, if the user answers "Tanaka," it verifies whether the answer is valid.

[1619] Notification of authentication results

[1620] The server determines whether the authentication was successful based on the user's answers and notifies the user terminal of the result. If the authentication was successful, a message to that effect is displayed on the terminal screen.

[1621] Specific examples

[1622] A specific example in which the user says "transfer money" will be shown below.

[1623] 1. User: Say "transfer money" into the terminal.

[1624] 2. Device: Records audio and sends the data to the server.

[1625] 3. Server: Convert the audio data into text using Google Cloud Speech-to-Text. Detect the phrase "transfer money."

[1626] 4. Server: Uses spaCy to analyze the context and scores the likelihood of fraud at 80%.

[1627] 5. Server: Use GPT-4 to generate the authentication question "What is your mother's maiden name?"

[1628] 6. Server: Sends the question to the user's terminal and displays it.

[1629] 7. User: Answers "Tanaka."

[1630] 8. Terminal: Send this response to the server.

[1631] 9. Server: Evaluate whether "Tanaka" is legitimate.

[1632] 10. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[1633] In this way, the system prevents fraud through advanced voice input analysis and authentication processes.

[1634] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1635] Step 1:

[1636] User: The user speaks into the terminal, saying "transfer money."

[1637] Specific action: Record audio using a microphone.

[1638] Input: User's voice

[1639] Output: Recorded audio data

[1640] Step 2:

[1641] Device: The device collects the recorded voice data and sends it to the server.

[1642] Specific operation: The recorded audio data is sent to the server via HTTPS.

[1643] Input: Recorded audio data

[1644] Output: Audio data sent to the server

[1645] Step 3:

[1646] Server: The server converts the received voice data into text data using a speech recognition module (e.g., Google Cloud Speech-to-Text).

[1647] Specific operation: Sends audio data to the API and receives it as text data.

[1648] Input: Audio data

[1649] Output: Text data

[1650] Step 4:

[1651] Server: The server uses a natural language processing library (e.g., spaCy) to detect whether specific keywords (e.g., "transfer money") are contained in the text data.

[1652] Specific operation: Matches text data with a keyword list and extracts matching keywords.

[1653] Input: Text data

[1654] Output: Text data containing keywords

[1655] Step 5:

[1656] Server: If a keyword is detected, the context is analyzed and a score is assigned for potential fraud.

[1657] What it does: Runs a contextual analysis algorithm and scores the likelihood of fraud on a scale of 0% to 100%.

[1658] Input: Text data containing keywords

[1659] Output: A score indicating the likelihood of fraud.

[1660] Step 6:

[1661] Server: If fraud is deemed likely, a generative AI model (e.g., GPT-4) is used to generate user authentication questions.

[1662] Specific operation: The AI ​​model is given a prompt, "Generate a question using information about the user's close relationships," and the generated question is received.

[1663] Input: Fraud likelihood score, prompt statement

[1664] Output: Generated authentication questions

[1665] Step 7:

[1666] Server: Sends the generated authentication question to the user's terminal and instructs it to be displayed.

[1667] Specific operation: The text data of the question is sent to the terminal via HTTPS.

[1668] Input: Generated authentication question

[1669] Output: Authentication question displayed on the user's terminal

[1670] Step 8:

[1671] User: The user enters the answer to the displayed authentication question. For example, the user answers "Tanaka."

[1672] Specific actions: Enter your answer in the input field on the device and press the send button.

[1673] Input: Answer to authentication question

[1674] Output: Answers typed into the terminal

[1675] Step 9:

[1676] Terminal: The terminal collects the user's answers and sends them to the server.

[1677] Specific operation: The entered answer is sent to the server via HTTPS.

[1678] Input: Answer typed into the terminal

[1679] Output: The answer sent to the server

[1680] Step 10:

[1681] Server: Compares the received user answers with a pre-configured database of correct answers and evaluates whether the user is legitimate.

[1682] Specific behavior: Compares the answer data with the correct answer database to see if it matches.

[1683] Input: User answers, correct answer database

[1684] Output: Evaluation result (valid or not)

[1685] Step 11:

[1686] Server: Generates the authentication result and notifies the user terminal.

[1687] Specific operation: Creates a message indicating whether authentication was successful or failed and sends it to the terminal via HTTPS communication.

[1688] Input: Evaluation result

[1689] Output: Authentication result sent to the terminal

[1690] Step 12:

[1691] Terminal: Displays the authentication result received from the server to the user.

[1692] Specific behavior: A message indicating authentication success or failure is displayed on the screen.

[1693] Input: Authentication result

[1694] Output: Authentication result displayed to the user

[1695] (Application example 1)

[1696] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1697] Current voice input systems have difficulty detecting fraud in real time and quickly authenticating users. Especially in electronic payments, where transactions are frequent and fast, there is a lack of means to prevent fraud immediately. Furthermore, fraud detection systems based on voice input often rely too much on specific keywords, making it difficult to accurately analyze the context, leading to false positives. Therefore, a system that can detect fraud and authenticate users with higher accuracy is needed.

[1698] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1699] In this invention, the server includes means for collecting voice input in real time, means for converting the voice input into text, means for detecting specific keywords in the text, means for analyzing the context and assessing the likelihood of fraud, means for generating additional authentication challenges if the likelihood of fraud is high, and means for displaying the generated challenges on the user terminal, thereby enabling instant fraud detection using voice and rapid user authentication.

[1700] "Voice input" refers to spoken content or instructions from a user through a voice collection device such as a microphone.

[1701] "Means for converting to text" refers to speech recognition technology or software for converting voice input into character string data.

[1702] "Means for detecting specific keywords" refers to the ability to identify and extract predefined important words and phrases within the converted text.

[1703] "Contextual fraud risk assessment" refers to algorithms that analyze the context of text and the use of language to score or assess the risk of fraud.

[1704] The "means for generating additional authentication questions" refers to the ability to create new questions to verify the user's authenticity if it is determined that there is a high possibility of fraud.

[1705] "Means for displaying on the user terminal" refers to the technology or interface for displaying the generated authentication question on the screen of the device used by the user.

[1706] "Means for recognizing voice in real time" refers to a function that instantly detects and recognizes the user's speech and then processes it.

[1707] "Means for obtaining analysis results on the server" refers to the function of obtaining the results of speech recognition and text analysis from a remote server and sharing and using them throughout the system.

[1708] A "generative AI model" is a machine learning model that uses artificial intelligence to analyze and process input data and generate appropriate output.

[1709] A "prompt" is a document containing instructions or questions to be input into a generative AI model, providing criteria and conditions for data analysis.

[1710] DETAILED DESCRIPTION OF THE INVENTION The following describes an embodiment of the present invention.

[1711] The server has a means for collecting voice input, and collects the voice spoken by the user into the terminal. The collected voice is converted into text using voice recognition technology. An example of the voice recognition technology used here is Google's voice recognition API. By using this API, the collected voice data can be converted into text in real time.

[1712] The converted text is sent to a server where it is processed to detect specific keywords. Natural Language Processing (NLP) techniques are recommended to detect specific keywords in the text. Once a specific keyword is detected, the server analyzes the context and assesses its likelihood of fraud. This assessment is performed using a generative AI model, which analyzes the prompt sentence as input.

[1713] If the server determines that fraud is likely, it generates additional questions for user authentication. These questions may contain information about the user's relatives, for example, and are important for verifying the user's authenticity. The generated questions are sent to the user's terminal and displayed on its screen.

[1714] The user answers the displayed questions, and the answers are sent to the server via the device. The server evaluates the user's answers and determines whether the user is legitimate. This authentication uses a generative AI model to instantly evaluate the validity of the answers.

[1715] Hardware or software used

[1716] Hardware: Smartphone, microphone

[1717] software:

[1718] Speech Recognition API (Google)

[1719] Natural Language Processing (NLP)

[1720] Generative AI Models

[1721] Libraries for handling HTTP requests (requests)

[1722] Realization of processing steps

[1723] For example, if the user says "I would like to purchase this item," the following steps are performed:

[1724] 1. Audio collection and conversion:

[1725] The user speaks into their smartphone, saying, "I would like to purchase this product." The device collects this speech and converts it into text using a speech recognition API.

[1726] 2. Text transmission and analysis:

[1727] The text is sent to a server, which detects specific keywords (such as "purchase") and analyzes the context. If it is determined to be fraudulent, the corresponding prompt is fed into a generative AI model.

[1728] 3. Generate and display authentication questions:

[1729] If the generative AI model scores a high probability of fraud, it generates additional authentication questions and displays them on the user's device, such as "What is your mother's maiden name?"

[1730] 4. User answers and their ratings:

[1731] If the user answers "Tanaka," the device sends the answer to the server, which evaluates the answer using a generative AI model to confirm its validity.

[1732] Prompt Sentence Examples

[1733] An example of a prompt used in this system is as follows:

[1734] "Please enter the following information into the system:

[1735] Say: "Buy this item."

[1736] Contextual analysis results: 80% probability of fraud.

[1737] Verification Question: What is your mother's maiden name?

[1738] User Answer: Tanaka.

[1739] Please evaluate the user for legitimacy and notify us of the results."

[1740] This embodiment allows for instant fraud detection and highly accurate user authentication based on voice input.

[1741] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1742] Step 1:

[1743] A user speaks into a device (smartphone) to input information. For example, the user might say, "I would like to purchase this product."

[1744] Input: User's voice data

[1745] Output: Collected audio data

[1746] Step 2:

[1747] The device sends the collected voice data in real time to a voice recognition API (for example, Google's voice recognition API), which converts the voice data into text data.

[1748] Input: Collected audio data

[1749] Output: Converted text data

[1750] Specific operation: Sends audio data to the API and receives text data from the API.

[1751] Step 3:

[1752] The device transmits the converted text data to a server, which performs text analysis to detect specific keywords in the text data, for example, the keyword "purchase."

[1753] Input: Text data

[1754] Output: Keyword detection results

[1755] What it does: Runs a text analysis algorithm to extract specific keywords.

[1756] Step 4:

[1757] The server analyzes the context of the detected keywords and utilizes a generative AI model to assess the likelihood of fraud. The prompt sentence is provided as input to the generative AI model, which scores the probability of fraud.

[1758] Input: Text data and detected keywords

[1759] Output: Fraud likelihood score

[1760] Specific operation: Contextual analysis is performed using NLP technology, prompt sentences are input into the generative AI model, and a score is calculated.

[1761] Step 5:

[1762] If the fraud score is high, the server generates additional questions to authenticate the user, for example, "What is your mother's maiden name?"

[1763] Input: Fraud likelihood score

[1764] Output: Generated authentication questions

[1765] What it does: Selects and generates appropriate questions from a generative AI model or a predefined question set.

[1766] Step 6:

[1767] The generated user authentication question is sent from the server to the terminal and displayed on the terminal screen.

[1768] Input: Generated authentication question

[1769] Output: Question displayed on terminal

[1770] Specific operation: Sends an HTTP request from the server to the device and executes the interface to display the question on the screen.

[1771] Step 7:

[1772] The user answers the authentication question displayed on the terminal. For example, the user answers "Tanaka."

[1773] Input: The question displayed and the user's answer

[1774] Output: User response data

[1775] What happens: The user enters their answers in text format into the device, and the device collects the data.

[1776] Step 8:

[1777] The device sends the collected user responses to the server, which evaluates the responses and uses a generative AI model to determine whether the user is legitimate.

[1778] Input: User response data

[1779] Output: User authentication evaluation result

[1780] What it does: The server analyzes the user's answers and evaluates them to ensure their validity using a generative AI model.

[1781] Step 9:

[1782] The server notifies the terminal of the evaluation result of the user authentication, and the terminal displays the result to the user. For example, if the authentication is successful, it displays "Authentication successful" to notify the user that suspicion of fraud has been cleared.

[1783] Input: User authentication evaluation result

[1784] Output: Notification message displayed on the terminal

[1785] Specific operation: The server sends the notification result to the device, and the device updates the UI elements to display the result to the user.

[1786] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1787] This invention combines an emotion engine with a system that detects potential fraud based on voice input and authenticates users as needed to achieve more accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[1788] System configuration and operation

[1789] 1. Collecting and converting voice input

[1790] The user speaks to the terminal, saying something like "transfer money." The terminal collects this voice input and sends it to the server, which then converts the received voice data into text using a voice recognition module.

[1791] 2. Keyword detection and context analysis

[1792] The server detects whether specific keywords (e.g., "transfer money") are contained in the converted text data. If a keyword is detected, the server performs contextual analysis, evaluates the surrounding context, and scores the likelihood of fraud.

[1793] 3. Emotion Recognition and Analysis

[1794] The server has an emotion engine for recognizing the user's emotions from the voice input. The emotion engine analyzes the tone, speed, accent, etc. of the voice to determine whether the user is in an emotional state such as anxiety, tension, excitement, etc.

[1795] 4. Generating and Displaying User Authentication Questions

[1796] If the server determines that fraud is likely or that the user is in a particular emotional state (e.g., anxiety or nervousness), it generates a user authentication challenge, which is specific to the individual user, such as information about the user's relatives. The challenge is then sent to the user's device and displayed.

[1797] 5. Collecting and Evaluating User Responses

[1798] The user answers the displayed authentication questions. The terminal collects the user's answers and sends them to the server. The server receives the answers and evaluates whether the user is legitimate. Based on this evaluation, the server determines whether the authentication succeeds or fails and notifies the terminal of the result.

[1799] Specific examples

[1800] The specific flow when the user says "transfer money" is shown below.

[1801] 1. User: Say "transfer money" into the terminal.

[1802] 2. Device: Records audio and sends the data to the server.

[1803] 3. Server: Converts the voice data into text and detects the phrase "transfer money."

[1804] 4. Server: Analyzes the context and scores the likelihood of fraud at 80%.

[1805] 5. Server: Uses an emotion engine to recognize whether the user is anxious or nervous.

[1806] 6. Server: Based on a comprehensive assessment of the likelihood of fraud and the emotional state, the server determines that the risk is high.

[1807] 7. Server: Generate the authentication question "What is your mother's maiden name?"

[1808] 8. Server: Sends the question to the user's terminal and displays it.

[1809] 9. User: Answers "Tanaka."

[1810] 10. Terminal: Send this response to the server.

[1811] 11. Server: Evaluate whether "Tanaka" is legitimate.

[1812] 12. Server: Notifies the device of the authentication result. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the device.

[1813] In this way, the system analyzes user comments and emotions through multiple steps to prevent fraud.

[1814] The processing flow will be explained below.

[1815] Step 1:

[1816] The user makes a statement such as "transfer money" through voice input.

[1817] Step 2:

[1818] The terminal records the user's voice in real time and saves it as voice data.

[1819] Step 3:

[1820] The terminal transmits the recorded voice data to the server.

[1821] Step 4:

[1822] The server converts the received voice data into text using a voice recognition module.

[1823] Step 5:

[1824] The server detects specific keywords (e.g., "transfer money") in the converted text.

[1825] Step 6:

[1826] When a specific keyword is detected, the server analyzes the surrounding context using a natural language processing (NLP) module.

[1827] Step 7:

[1828] The server scores the likelihood of fraud based on the results of the contextual analysis and determines when the risk of fraud is high.

[1829] Step 8:

[1830] The server analyzes the voice data using an emotion engine to recognize the user's emotional state (e.g., anxiety, tension, excitement).

[1831] Step 9:

[1832] The server evaluates whether the user is in a particular emotional state based on the output of the emotion engine.

[1833] Step 10:

[1834] The server comprehensively assesses the fraud risk and the emotional state and determines that there is a high possibility of fraudulent activity.

[1835] Step 11:

[1836] The server generates a user authentication question, such as "What is your mother's maiden name?"

[1837] Step 12:

[1838] The server sends the generated authentication challenge to the user's terminal.

[1839] Step 13:

[1840] The terminal displays the authentication challenge received from the server to the user.

[1841] Step 14:

[1842] The user answers the displayed authentication question (e.g., answers "Tanaka").

[1843] Step 15:

[1844] The terminal transmits the user's answer to the server.

[1845] Step 16:

[1846] The server evaluates the received user answers and verifies their validity.

[1847] Step 17:

[1848] The server notifies the user terminal of the authentication result. If the user is legitimate, it sends a "Authentication successful" message, otherwise it sends a "Authentication failed" message.

[1849] Example 2

[1850] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1851] Conventional systems have limitations in the accuracy of fraud detection based on voice input, resulting in the possibility of false positives and false negatives. Furthermore, because they do not evaluate the user's emotional state, they are unable to respond to situations where an emotional approach would be expected to improve accuracy. This creates the risk of overlooking sophisticated fraud.

[1852] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1853] In this invention, the server includes means for collecting voice input, means for converting the collected voice input into text, means for detecting specific keywords in the text, means for analyzing the context surrounding the detected keywords and assessing the possibility of fraud, means for generating questions for user authentication when there is a high possibility of fraud, means for displaying the generated questions, means for collecting user answers, means for evaluating the collected answers and determining whether the user is legitimate, means for recognizing the user's emotions from the voice and assessing the user's emotional state, and means for comprehensively assessing the possibility of fraud and the emotional state. This enables highly accurate detection of fraud and user authentication by analyzing both the voice input and the emotional state.

[1854] "Voice input" refers to the voice that the user speaks into the terminal, and this voice is the voice data collected by the system.

[1855] "Means of collection" refers to the function of capturing voice data using a microphone or sensor built into the device.

[1856] "Means for converting to text" refers to the function of converting collected voice data into text data using voice recognition technology.

[1857] "Specific Keywords" refers to specific words or phrases that the system considers important for fraud detection and contextual analysis.

[1858] "Means for analyzing context" refers to the function of using natural language processing technology to understand the sentences before and after specific keywords and evaluate their meaning and relevance.

[1859] "Means for assessing the likelihood of fraud" refers to a function that assesses the risk level of fraud as a number or score based on the results of contextual analysis.

[1860] "User authentication questions" refer to questions generated to verify a user's identity when fraud is deemed likely.

[1861] "Means for generating" refers to algorithms or models for automatically creating user authentication questions, in particular generative AI models.

[1862] The "display means" refers to a function for displaying the generated question on the user's terminal screen.

[1863] "Means for collecting answers" refers to the functionality for incorporating answers provided by users into the system.

[1864] "Means for evaluation" refers to the function of analyzing collected responses and determining whether they are from legitimate users.

[1865] "Means for recognizing emotions" refers to technology for analyzing a user's emotional state from voice data, evaluating the tone, speed, accent, etc. of the voice.

[1866] The "means for evaluating emotional state" refers to a function for evaluating the user's psychological state as a numerical value or a category based on the recognized emotional data.

[1867] "Means for making a comprehensive judgment" refers to the function of taking into consideration both the assessment of the possibility of fraud and the assessment of emotional state to make a final judgment.

[1868] This invention is a system that detects possible fraudulent activity based on voice input and authenticates users as necessary. Furthermore, this system is combined with an emotion engine to achieve highly accurate fraud detection and user authentication. This system is comprised of a combination of speech recognition technology, natural language processing (NLP), and emotion recognition technology.

[1869] System Configuration and Operation

[1870] Hardware and Software Overview

[1871] 1. Collecting voice input

[1872] The user uses the built-in microphone of the terminal to input voice commands such as "transfer money."

[1873] The terminal transmits the collected voice data to the server.

[1874] The server converts the audio data into text using the Google Cloud Speech-to-Text API.

[1875] 2. Keyword detection and context analysis

[1876] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data.

[1877] If keywords are detected, the server performs a contextual analysis using natural language processing techniques (e.g., spaCy) to assess the likelihood of fraud.

[1878] The likelihood of fraud is scored, for example, rated as 80%.

[1879] 3. Emotion recognition

[1880] The server uses an emotion recognition engine such as DeepAffects to recognize the user's emotions from their voice.

[1881] It analyzes the tone, speed and accent of the voice to determine whether the user is anxious or nervous.

[1882] 4. Generating and displaying user authentication questions

[1883] If fraud is deemed likely or the user is in a particular emotional state, the server uses a generative AI model (e.g., GPT-3) to generate user authentication questions.

[1884] Sample question: "What is your mother's maiden name?"

[1885] The server sends the generated question to the user's terminal, where it is displayed.

[1886] 5. Collecting and Evaluating User Answers

[1887] The user answers the displayed question, for example, by replying "Tanaka."

[1888] The terminal sends this response to the server.

[1889] The server analyzes the response and checks the database to see if the user is legitimate.

[1890] The authentication result is notified to the terminal, and if the suspicion of fraud is cleared, a message indicating successful authentication is displayed on the terminal.

[1891] Specific examples

[1892] When the user says "transfer money", the system behaves as follows:

[1893] 1. User: Say "transfer money" into the terminal.

[1894] 2. Device: Records audio using the built-in microphone and sends the audio data to the server.

[1895] 3. Server: Convert the audio data into text using the Google Cloud Speech-to-Text API.

[1896] 4. Server: Detects the keyword "transfer money" from the converted text.

[1897] 5. Server: Analyze the context with spaCy and score the likelihood of fraud at 80%.

[1898] 6. Server: Analyze the user's emotional state from their voice using DeepAffects. Recognize emotions such as anxiety and tension.

[1899] 7. Server: Use the GPT-3 model to generate the authentication question, "What is your mother's maiden name?"

[1900] 8. Server: Sends the question to the user's device and displays it on the device.

[1901] 9. User: Answers "Tanaka."

[1902] 10. Terminal: Send this response to the server.

[1903] 11. Server: Evaluates the answers against a database.

[1904] 12. Server: The authentication result is notified to the terminal, and if there is no suspicion of fraud, a message indicating successful authentication is displayed.

[1905] The system enables highly accurate fraud detection and user authentication through voice and emotion analysis.

[1906] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1907] Step 1:

[1908] The user uses the built-in microphone of the device to input voice commands such as "transfer money." This becomes the input data (voice data).

[1909] Step 2:

[1910] The device records the voice data collected by the built-in microphone. Then, this voice data is sent to the server, and the sent voice data becomes the input data for the server.

[1911] Step 3:

[1912] The server converts the received voice data into text data using a voice recognition module (e.g., Google Cloud Speech-to-Text API). The input is voice data, and the output is text data.

[1913] Step 4:

[1914] The server detects whether a specific keyword (e.g., "transfer money") is included in the converted text data. The input is the text data, and the output is the keyword detection results and their location information.

[1915] Step 5:

[1916] If a keyword is detected, the server uses natural language processing (NLP) technology (e.g., spaCy) to analyze the context and evaluate the likelihood of fraud. The input is text data and keyword detection results, and the output is a score indicating the likelihood of fraud.

[1917] Step 6:

[1918] The server uses emotion recognition technology (e.g., DeepAffects) to analyze the user's emotions from the voice data. The input is the voice data, and the output is the evaluation result of the emotional state. Specifically, it analyzes voice characteristics such as tone, speed, and accent.

[1919] Step 7:

[1920] The server comprehensively assesses the fraud likelihood assessment and emotional state assessment, and if it determines that the risk is high, it uses a generative AI model (e.g., GPT-3) to generate questions for user authentication. The input is the fraud likelihood score and the emotional state assessment result, and the output is questions for user authentication. Questions that are generated include, "What is your mother's maiden name?"

[1921] Step 8:

[1922] The server sends the generated question to the user's terminal, and the terminal displays the question on the screen. The input is the question for user authentication, and the output is the question displayed on the terminal.

[1923] Step 9:

[1924] The user answers the question displayed on the terminal. For example, the answer "Tanaka" is used. This becomes the input data.

[1925] Step 10:

[1926] The terminal collects the user's answers and sends them to the server. The input is the user's answers, and the output is the answer data sent to the server.

[1927] Step 11:

[1928] The server analyzes the received response data and compares it with a database to determine whether the user is a legitimate user. The input is the user's response data, and the output is the authentication result.

[1929] Step 12:

[1930] The server notifies the user's terminal of the authentication result. The terminal displays this authentication result to the user. The input is the authentication result, and the output is the authentication result displayed on the terminal. If there is no suspicion of fraud, a message indicating successful authentication is displayed on the terminal.

[1931] (Application example 2)

[1932] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1933] Conventional fraud detection systems rely solely on simple keyword detection using voice recognition, resulting in frequent inaccuracies. Furthermore, they fail to consider the user's emotional state, resulting in insufficient fraud risk assessment. The present invention aims to address these issues and achieve more accurate fraud detection and user authentication.

[1934] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 2 is realized by the following means.

[1935] In this invention, the server includes means for collecting voice input, means for analyzing emotional state from the voice data, and means for generating user authentication challenges when fraud is likely, thereby taking into account the user's emotional state and providing real-time warnings of potential fraud, as well as providing immediate additional user authentication if necessary.

[1936] A "means for collecting voice input" is a device or software capable of recording user-uttered speech on the device and transmitting it to a server for analysis.

[1937] A "means for converting collected voice input to text" is a process or device that uses voice recognition technology to convert voice data into written data.

[1938] The "means for detecting specific keywords in text" refers to a device or software that has the function of identifying and extracting predetermined keywords from the converted text data.

[1939] "Means for analyzing the context surrounding detected keywords and assessing the likelihood of fraud" refers to a process or device that performs a detailed analysis of the context surrounding keywords and assesses the risk of fraud based on that content.

[1940] A "means for generating user authentication questions when fraud is likely" is a process or device that automatically generates additional user verification questions when a high risk of fraud is determined.

[1941] The "means for displaying the generated question" is a device or software that provides an interface for displaying the generated authentication question on the user's terminal.

[1942] The "means for collecting user answers" is a process or device that collects the answers given by the user to the authentication questions and transmits them to the server.

[1943] The "means for evaluating collected answers and determining whether a user is legitimate" is a process or device that analyzes collected answers and evaluates their legitimacy.

[1944] The "means for analyzing emotional state from voice data" is a process or device that analyzes the tone, rate, accent, etc. of the voice to determine the emotional state of the user.

[1945] A "means for alerting a user to the possibility of fraud in real time while the user is on a call" is a device or software that has the function of assessing the risk of fraud in real time while the user is on a call and immediately notifying the user of the results.

[1946] This invention is a system that detects possible fraud based on voice input and authenticates users as needed, and by combining it with an emotion engine, achieves more accurate fraud detection and user authentication. The invention is implemented by a combination of specific hardware and software.

[1947] Hardware and software used

[1948] 1. Smartphones: Collect voice input.

[1949] 2. Speech recognition software (e.g. speech_recognition package): converts voice data into text.

[1950] 3. Emotion recognition API (e.g. emotion-detect API): Analyzes the user's emotional state.

[1951] 4. Natural language processing engine: Detects specific keywords from text data and analyzes the context.

[1952] Processing flow

[1953] 1. A means of converting collected voice input into text

[1954] Voice data sent by a user into a smartphone is converted into text data using voice recognition software.

[1955] 2. How to find specific keywords

[1956] Detect specified keywords (e.g., "transfer," "remittance," "money") from the converted text data.

[1957] 3. A means of analyzing context and assessing potential fraud

[1958] The context surrounding the detected keywords is analyzed using a natural language processing engine to score the likelihood of fraud.

[1959] 4. A means of analyzing emotional states

[1960] The collected voice data is sent to an emotion recognition API to analyze the user's emotional state (e.g., anxiety, tension).

[1961] 5. Generating User Authentication Challenges When Fraud is Highly Likely

[1962] If fraud is deemed likely, the system generates a user authentication challenge based on the user's personal information, taking into account the user's emotional state.

[1963] 6. A way to display the generated questions

[1964] The generated question is displayed on the smartphone, prompting the user to answer.

[1965] 7. How to collect user responses

[1966] The user's answers are collected and sent to the server.

[1967] 8. A means of evaluating collected responses and determining whether the user is legitimate

[1968] The server evaluates the user's answer and determines whether it is valid.

[1969] Specific examples

[1970] For example, if a user says "transfer money," the voice input is recorded and converted to text using speech recognition software. Corresponding keywords are then detected, and context and sentiment analysis is performed. If the server determines that the user is anxious or nervous, along with the possibility of fraud, it generates an additional authentication question, "What is your mother's maiden name?", which is displayed on the user's smartphone. If the user enters the answer "Tanaka," the answer is sent to the server and evaluated for validity.

[1971] Prompt Sentence Examples

[1972] "Analyze the voice of someone saying they're transferring money, analyze the emotion, and if there's a possibility of fraud, generate additional authentication questions for the user."

[1973] As such, the present invention is a system that combines voice recognition, natural language processing, and emotion analysis to achieve highly accurate fraud detection and user authentication in real time.

[1974] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1975] Step 1:

[1976] The user collects voice input.

[1977] Input: Voice data spoken by the user into the smartphone.

[1978] What happens: The user says, "Transfer money." The smartphone's microphone records this speech and generates a digital audio file.

[1979] Output: Collected audio data.

[1980] Step 2:

[1981] The device sends the voice data to a server and uses a means to convert the voice input into text.

[1982] Input: Audio data.

[1983] Specific operation: The smartphone sends the recorded voice data to the server, where it is converted into text data using voice recognition software (e.g., Google Speech Recognition API).

[1984] Output: Text data.

[1985] Step 3:

[1986] The server uses a means for detecting specific keywords from the text data.

[1987] Input: Text data.

[1988] Specific operation: The server's natural language processing engine searches and detects keywords such as "transfer," "remittance," and "money" within the text data.

[1989] Output: The detected keywords.

[1990] Step 4:

[1991] The server uses a method to analyze the context of the keywords and assess the likelihood of fraud.

[1992] Input: The detected keyword and its surrounding text context.

[1993] How it works: The natural language processing engine analyzes the context surrounding the keyword and scores the likelihood of fraud based on that content. For example, it evaluates the rationality of the action of "transferring money" from the context.

[1994] Output: A score indicating the likelihood of fraud.

[1995] Step 5:

[1996] The server uses a means for analyzing the emotional state from the voice data.

[1997] Input: Audio data.

[1998] Specific operation: The server sends the voice data to an emotion recognition API (e.g., emotion-detect API) and analyzes the emotional state (e.g., anxiety, tension).

[1999] Output: User's emotional state data.

[2000] Step 6:

[2001] The server uses a means to assess the likelihood of fraud and emotional state and generate a user authentication challenge if fraud is likely.

[2002] Input: Deception score, user emotional state data.

[2003] What it does: The server evaluates the fraud score and emotional state data together, and generates user authentication questions (e.g., "What is your mother's maiden name?") if the risk is deemed high.

[2004] Output: The generated authentication questions.

[2005] Step 7:

[2006] The terminal uses a means for displaying the generated question.

[2007] Input: The generated authentication question.

[2008] Specific operation: The server generates a question, which is sent to the smartphone and displayed on the smartphone screen.

[2009] Output: A smartphone screen showing the authentication questions.

[2010] Step 8:

[2011] The user provides answers to questions using means collected by the terminal.

[2012] Input: The user's answer.

[2013] Specific operation: The user enters an answer (e.g., "Tanaka") to the displayed question, and the smartphone collects the answer and sends it to the server.

[2014] Output: Collected response data.

[2015] Step 9:

[2016] The server evaluates the collected answers and uses a means to determine if the user is legitimate.

[2017] Input: Collected response data.

[2018] Specific operation: The server checks the user's answer against a pre-registered database to evaluate its validity. If the answer is valid, it sends a message to the smartphone indicating successful authentication.

[2019] Output: Data indicating the success or failure of the authentication, and a message based on the result.

[2020] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[2021] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[2022] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[2023] The emotion identification model 59 as an emotion engine may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[2024] FIG. 9 is a diagram illustrating an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and actions arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[2025] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[2026] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[2027] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[2028] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[2029] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[2030] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[2031] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[2032] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[2033] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[2034] It is not necessary to store all of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store all of the specific processing program 56 in the storage 32; only a portion of the specific processing program 56 may be stored.

[2035] The hardware resource for executing a specific process can be any of the following processors: An example of a processor is a CPU, which is a general-purpose processor that functions as a hardware resource for executing a specific process by executing software, i.e., a program. Another example of a processor is a dedicated electrical circuit, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically for executing a specific process. Each processor has built-in or connected memory, and each processor uses the memory to execute the specific process.

[2036] The hardware resource that executes the specific processing may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Also, the hardware resource that executes the specific processing may be a single processor.

[2037] As an example of a system configured with a single processor, first, one processor is configured by combining one or more CPUs and software, and this processor functions as a hardware resource that executes a specific process. Second, there is a system that uses a processor that realizes the functions of an entire system including multiple hardware resources that execute a specific process on a single IC chip, as typified by SoC (System-on-a-chip). In this way, a specific process is realized using one or more of the above-mentioned various processors as hardware resources.

[2038] Furthermore, the hardware structure of these various processors can be, more specifically, an electric circuit that combines circuit elements such as semiconductor devices. The specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps may be deleted, new steps may be added, or the processing order may be rearranged, without departing from the spirit of the invention.

[2039] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[2040] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[2041] The following is further disclosed regarding the above embodiment.

[2042] (Claim 1)

[2043] means for collecting voice input;

[2044] means for converting the collected voice input into text;

[2045] means for detecting specific keywords within the text;

[2046] a means for analyzing the context of the detected keywords to assess the likelihood of fraud;

[2047] means for generating user authentication challenges when fraud is likely;

[2048] a means for displaying the generated questions;

[2049] a means for collecting user responses;

[2050] a means for evaluating the collected responses to determine whether the user is legitimate; and

[2051] A system including:

[2052] (Claim 2)

[2053] 2. The system of claim 1, wherein the user authentication challenge includes information about the user's relatives.

[2054] (Claim 3)

[2055] 10. The system of claim 1, further comprising means for scoring the likelihood of fraud.

[2056] "Example 1"

[2057] (Claim 1)

[2058] means for collecting voice input;

[2059] means for converting the collected voice input into text;

[2060] means for detecting specific keywords within the text;

[2061] a means for analyzing the context of the detected keywords to assess the likelihood of fraud;

[2062] means for generating user authentication challenges when fraud is likely;

[2063] means for generating user authentication questions using the generative AI model;

[2064] means for transmitting the generated question to a user terminal and displaying it;

[2065] a means for collecting user responses;

[2066] a means for evaluating the collected responses to determine whether the user is legitimate; and

[2067] means for notifying the user terminal of the authentication result;

[2068] A system including:

[2069] (Claim 2)

[2070] 10. The system of claim 1, wherein the user authentication challenge includes information about the user's close relationships.

[2071] (Claim 3)

[2072] 10. The system of claim 1, further comprising means for scoring the likelihood of fraud.

[2073] "Application Example 1"

[2074] (Claim 1)

[2075] means for collecting voice input;

[2076] means for converting the collected voice input into text;

[2077] means for detecting specific keywords within the text;

[2078] a means for analyzing the context of the detected keywords to assess the likelihood of fraud;

[2079] means for generating user authentication challenges when fraud is likely;

[2080] a means for displaying the generated questions;

[2081] a means for collecting user responses;

[2082] a means for evaluating the collected responses to determine whether the user is legitimate; and

[2083] means for generating and displaying additional authentication questions on the user terminal after the likelihood of fraud has been assessed;

[2084] A means of recognizing voice in real time and obtaining the analysis results on the server,

[2085] A system including:

[2086] (Claim 2)

[2087] 2. The system of claim 1, wherein the user authentication challenge includes information about the user's relatives.

[2088] (Claim 3)

[2089] 10. The system of claim 1, further comprising means for scoring the likelihood of fraud.

[2090] (Claim 4)

[2091] 2. The system of claim 1, wherein the step of obtaining the analysis results from the server is related to the user's electronic payment activity.

[2092] (Claim 5)

[2093] The system of claim 1 further comprising means for processing the speech recognition and analysis results using a generative AI model based on a prompt sentence.

[2094] "Example 2: Combining Emotion Engines"

[2095] (Claim 1)

[2096] means for collecting voice input;

[2097] means for converting the collected voice input into text;

[2098] means for detecting specific keywords within the text;

[2099] a means for analyzing the context of the detected keywords to assess the likelihood of fraud;

[2100] means for generating user authentication challenges when fraud is likely;

[2101] a means for displaying the generated questions;

[2102] a means for collecting user responses;

[2103] a means for evaluating the collected responses to determine whether the user is legitimate; and

[2104] means for recognizing a user's emotion from speech and assessing the user's emotional state;

[2105] A means for comprehensively assessing the likelihood of fraud and the emotional state;

[2106] A system including:

[2107] (Claim 2)

[2108] 2. The system of claim 1, wherein the user authentication challenge includes information about the user's relatives.

[2109] (Claim 3)

[2110] 10. The system of claim 1, further comprising means for scoring the likelihood of fraud.

[2111] "Application example 2 when combining emotion engines"

[2112] (Claim 1)

[2113] means for collecting voice input;

[2114] means for converting the collected voice input into text;

[2115] means for detecting specific keywords within the text;

[2116] a means for analyzing the context of the detected keywords to assess the likelihood of fraud;

[2117] means for generating user authentication challenges when fraud is likely;

[2118] a means for displaying the generated questions;

[2119] a means for collecting user responses;

[2120] a means for evaluating the collected responses to determine whether the user is legitimate; and

[2121] means for analyzing emotional state from audio data;

[2122] a means for alerting users to potential fraud in real time while they are on a call;

[2123] A system including:

[2124] (Claim 2)

[2125] 10. The system of claim 1, wherein the user authentication challenge includes information about the user's relatives and is generated in real time for additional authentication.

[2126] (Claim 3)

[2127] 10. The system of claim 1, further comprising means for scoring the likelihood of fraud and means for assessing risk based on emotion recognition results. [Explanation of symbols]

[2128] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. means for collecting voice input; means for converting the collected voice input into text; means for detecting specific keywords within the text; a means for analyzing the context of the detected keywords to assess the likelihood of fraud; means for generating user authentication challenges when fraud is likely; a means for displaying the generated questions; a means for collecting user responses; a means for evaluating the collected responses to determine whether the user is legitimate; and A system including:

2. 2. The system of claim 1, wherein the user authentication challenge includes information about the user's relatives.

3. The system of claim 1 further comprising means for scoring the likelihood of fraud.

Images