On-device-based personal data leakage prevention and personalization response systems and methods

Abstract

An on-device-based personal information leakage prevention and personalized response system according to one embodiment of the present invention includes a plurality of user terminals that detect PII (Personal Identifiable Information) from input user questions and transfer user-neutral questions that convert the PII into neutral information, and a management server that receives the user-neutral questions and trains a management language model that generates common response patterns for each neutral question pattern.

Description

Technical Field

[0001] The present invention relates to a system and method for preventing the leakage of on-device-based personal information and for personalized response.

Background Art

[0002] Recently, in a situation where generative dialogue artificial intelligence such as ChatGPT has attracted great attention, concerns about the leakage of personal identification information (Personal Identifiable Information (hereinafter, PII)) of artificial intelligence and the limitation of the absence of a personalization function have been revealed.

[0003] In the case of existing smartphones, only a very small part of PII, such as fingerprint and face recognition information, is not transmitted to the management server and is processed within the user's terminal. On the other hand, in the case of recently newly emerged generative dialogue AI, the amount of PII that needs to be stored is much larger, and models that can answer questions not only through text question and answer like ChatGPT but also through images and videos with much larger data volumes have emerged. When a user uses a personal device for a long time, there is a limit to the existing method of managing only a very small part of PII in the device in terms of the further increase in the amount of accumulated information.

[0004] On the other hand, personalization of dialogue AI is always necessary to provide responses that match the user's tendencies, but there is a problem that technology for preventing the leakage of PII must be supplemented in this process.

[0005] Therefore, in order to solve the above problems, the present invention proposes an artificial intelligence agent that operates on a user terminal such as a smartphone without transmitting PII to a management server such as the cloud.

Summary of the Invention

Problems to be Solved by the Invention

[0006] The present invention aims to solve the aforementioned problems and to provide an on-device-based personal information leakage prevention and personal response system and method in which PII exists only on the user terminal and is not leaked externally, and generates personalized responses through a program inside the user terminal.

[0007] However, the technical challenges that this embodiment aims to address are not limited to those described above, and other technical challenges may exist. [Means for solving the problem]

[0008] As a technical means to solve the technical problems described above, the first aspect of the present invention, an on-device-based system for preventing the leakage of personal information and a personalized response system, includes multiple user terminals that detect PII (Personally Identifiable Information) from input user questions and transmit user-neutral questions converted from PII into neutral information, and a management server that receives user-neutral questions and trains a management language model to generate a common response pattern for each neutral question pattern.

[0009] The second aspect of the present invention, which provides on-device-based prevention of personal information leakage and a personalized response user terminal, includes a communication module, a memory storing a personalized response program, and a processor that executes the personalized response program. The personalized response program detects PII from the input user question, converts the PII into neutral information, and transfers the converted user-neutral question to a management server.

[0010] A third aspect of the present invention relates to the prevention of on-device-based leakage of personal information and a personal response method performed by a personal response system, which includes (a) a management server receiving user-neutral questions from multiple user terminals in which PII contained in user questions has been converted into neutral information, and (b) a management server receiving user-neutral questions and training a management language model that generates a common response pattern for each neutral question pattern. [Effects of the Invention]

[0011] According to any of the aforementioned solutions to the problems of the present invention, the PII exists only on the user terminal and is not leaked externally, and personalized responses can be generated through a program inside the device. Furthermore, a personalized AI question-and-answer service can be provided. [Brief explanation of the drawing]

[0012] [Figure 1] This is a diagram illustrating the configuration of an on-device-based personal information leakage prevention and personal response system according to one embodiment of the present invention. [Figure 2] This diagram illustrates an on-device-based personal information leakage prevention and personalized response system according to one embodiment of the present invention. [Figure 3] This figure illustrates an example of PII according to one embodiment of the present invention. [Figure 4] This diagram shows the configuration of an on-device-based personal information leakage prevention and a personalized response user terminal according to another embodiment of the present invention. [Figure 5] This is a flowchart illustrating a method for preventing the leakage of personal information and providing personalized responses, as performed on a user terminal according to another embodiment of the present invention. [Figure 6] This figure illustrates a personalized response program executed on a user terminal according to another embodiment of the present invention. [Figure 7] This figure illustrates the data interaction between a management server and a user terminal according to another embodiment of the present invention. [Figure 8] This flowchart illustrates another embodiment of the present invention that demonstrates on-device-based prevention of personal information leakage and a personalized response method performed by a personalized response system. [Modes for carrying out the invention]

[0013] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, so that they can be easily implemented by a person with ordinary skill in the art to which the present invention pertains. However, the present invention can be embodied in a variety of different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly illustrate the present invention with reference to the drawings, parts unrelated to the description have been omitted, and similar parts throughout the specification have been given similar reference numerals.

[0014] Throughout the specification, "connected" to another part includes not only "directly connected" parts but also "electrically connected" parts with other elements in between. Furthermore, "contains" a component, unless otherwise stated, means that it may contain other components rather than excluding them.

[0015] In this specification, “Unit” includes units implemented by hardware, units implemented by software, and units implemented using both. Furthermore, one unit may be implemented using two or more hardware components, and two or more units may be implemented by one hardware component. On the other hand, “~Unit” is not limited to software or hardware; “~Unit” may be configured to reside in an addressable storage medium, or to regenerate one or more processors. Therefore, as an example, “~Unit” includes components such as software components, object-oriented software components, class components, and task components, processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. Components and the functions provided within “~Unit” can be combined with fewer components and “~Unit” or further separated into additional components and “~Unit”. Moreover, components and “~Unit” can be embodied to regenerate one or more CPUs within a device.

[0016] A network refers to a connection structure that enables information exchange among each node such as terminals and servers, including local area networks (LANs), wide area networks (WANs), the Internet (WWW), wired and wireless data communication networks, telephone networks, wired and wireless television communication networks, etc. Examples of wireless data communication networks include, but are not limited to, 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), LTE (Long Term Evolution), WiMAX (World Interoperability for Microwave Access), Wi-Fi, Bluetooth (registered trademark) communication, infrared communication, ultrasonic communication, visible light communication (VLC), LiFi, etc.

[0017] FIG. 1 is a configuration diagram of an on-device-based personal information leakage prevention and personalization response system according to an embodiment of the present invention, and FIG. 2 is a diagram for explaining the on-device-based personal information leakage prevention and personalization response system according to an embodiment of the present invention.

[0018] Referring to FIG. 1, the personalization response system 1 can include a plurality of user terminals 10, a management server 20, and a database 30.

[0019] Referring to FIG. 2, the user terminal 10 can detect PII (Personal Identifiable Information) from the input user query and transfer the user-neutral query obtained by converting the PII into neutral information to the management server 20. At this time, PII means various information that can be used to reveal the identity of a specific individual, such as a resident registration number, name, email address, or the like.

[0020] The management server 20 can cause the management language model 210, which receives user-neutral questions from a plurality of user terminals 10 and generates a common response pattern for each neutral question pattern, to learn.

[0021] The database 30 can store or provide, as learning data, user-neutral questions and response data generated by the interaction between the management server 20 and the user terminals 10. As an example, the learning data managed by the database 30 can be used for federated learning in which models learn in cooperation with each other without directly sharing data stored in a plurality of locations.

[0022] The management server 20 can be implemented by a computer or a mobile terminal connectable to a network. Here, the computer includes, for example, a notebook computer, a desktop, a laptop, etc., and the mobile terminal can include any type of handheld-based wireless communication device such as various smartphones, tablet PCs, smartwatches, etc., as a wireless communication device that guarantees portability and mobility.

[0023] Also, the management server 20 can provide the user terminals 10 with a management language model learned based on user-neutral questions. At this time, the management server 20 can operate in a cloud computing service model such as SaaS (Software as a Service), PaaS (Platform as a Service), or IaaS (Infrastructure as a Service), or can be constructed in a form such as a private cloud, a public cloud, or a hybrid cloud.

[0024] Specifically, the management server 20 includes a communication module, memory, and a processor. The communication module provides a communication interface necessary to provide signals to be sent and received to the user terminal 10 in the form of packet data in conjunction with the communication network. Here, the communication module may be a device that includes hardware and software necessary to send and receive signals such as control signals or data signals with other network devices via wired or wireless connections.

[0025] The memory stores personalized response programs. The memory also performs the function of temporarily or permanently storing data processed by the processor. Here, the memory may include volatile storage media or non-volatile storage media, but the scope of the present invention is not limited to these.

[0026] Memory can also store other programs, such as operating systems for processing and controlling the processor, and can perform functions for the temporary storage of input or output data.

[0027] Memory may include at least one type of storage medium from among flash memory type, hard disk type, multimedia card micro type, card type memory (e.g., SD or XD memory), RAM, and ROM.

[0028] The processor executes personalized response programs and provides the ability to control the terminal hardware in response to program execution. Specifically, the processor can perform hardware control functions such as necessary file systems, memory allocation, networking, basic libraries, timers, device control (display, media, input devices, 3D, etc.), and other utilities in response to program execution.

[0029] Here, "processor" can include any kind of device capable of processing data, such as a processor. Here, "processor" can mean a data processing device built into hardware, for example, that has physically structured circuitry to perform functions expressed by code or commands contained within a program. Examples of such data processing devices built into hardware include microprocessors, central processing units (CPUs), processor cores, multiprocessors, ASICs (application-specific integrated circuits), and FPGAs (field programmable gate arrays), but the scope of the present invention is not limited to these.

[0030] A personalized response program according to one embodiment of the present invention can input user-neutral questions received from multiple user terminals 10 into a management language model 210 and generate a common response pattern for each neutral question pattern. For example, the management language model 210 can be updated through federated continual learning to derive a common response pattern that is commonly useful for a large number of users. Subsequently, information on the update of the management language model 210 can be transferred to each user terminal 10 and used to update the response generation model 132 in each user terminal 10.

[0031] For example, the managed language model 210 can extract sentence structure elements, divided into words, phrases, and clauses, from the text input in a user-neutral question, classify the sentence structure elements into pre-defined neutral question patterns, and learn by matching them with common response patterns corresponding to the neutral question patterns. For example, the neutral question patterns may be classified for each context based on a language distribution that includes the frequency and type of words in the user's neutral question.

[0032] The managed language model 210 can learn PII patterns for each neutral question pattern by utilizing the position of neutral information in the sentence structure for user-neutral questions. In this case, the neutral information is such that the PII included in the user question is a representative neutral word or <mask>This means that it has been converted to a neutral word or <mask>By understanding the location of tokens, we can determine the location of neutral information.

[0033] For example, each user terminal 10 can generate various patterns containing PII in user questions through dialogue. As an example, the first user can generate a pattern (first PII pattern) in which PII is included at position A in a neutral question pattern with the sentence structure "ABC". As another example, the second user can generate a pattern (second PII pattern) in which PII is included at position C in a neutral question pattern with the sentence structure "ABC". In other words, depending on the user's situation, the same word may be judged differently as PII for each user. For example, "Google" may be a simple word for the first user, but it may be PII as workplace information for the second user.

[0034] Therefore, the management server 20 can determine, for each neutral question pattern received from multiple user terminals 10, which PII (Personal Information Indicator) was converted into neutral information in what context.

[0035] Figure 3 illustrates an example of PII according to one embodiment of the present invention.

[0036] Referring to Figure 3, PII can be divided into direct identifiers and quasi-identifiers that can identify a specific individual. Exemplary, direct identifiers refer to information in a PII dataset that can directly identify a specific individual on its own. For example, direct identifiers may include names, resident registration numbers, addresses, telephone numbers, and email addresses. Quasi-identifiers, on the other hand, refer to information that can indirectly identify a specific individual when combined with other quasi-identifiers. For example, quasi-identifiers may include age, gender, political leanings, religion, and customs. Methods for detecting PII will be described later with reference to Figures 5 through 7.

[0037] Figure 4 is a diagram illustrating the configuration of an on-device-based personal information leakage prevention and personalized response user terminal according to another embodiment of the present invention; Figure 5 is a flowchart illustrating the personal information leakage prevention and personalized response method performed on a user terminal according to another embodiment of the present invention; Figure 6 is a diagram illustrating a personalized response program executed on a user terminal according to another embodiment of the present invention; and Figure 7 is a diagram illustrating the data interaction between a management server and a user terminal according to another embodiment of the present invention.

[0038] Referring to Figure 4, the user terminal 10 may include a communication module 110, memory 120, processor 130, and database 140.

[0039] The communication module 110 can receive updated management language model information from the management server 20 and transfer it to the processor 130. Here, the communication module 110 may be a device that includes hardware and software necessary for sending and receiving signals such as control signals or data signals with other network devices via wired or wireless connections.

[0040] Memory 120 may contain a personalized response program. The personalized response program detects PII from the input user question, converts the PII into neutral information, and transfers the converted user-neutral question to the management server. Here, memory 120 may include a magnetic storage medium or a flash storage medium in addition to a volatile storage device that requires power to maintain the stored information, but the scope of the present invention is not limited to these.

[0041] Memory 120 can also store other programs, such as an operating system for processing and controlling the processor 130, and can perform functions for temporarily storing input or output data.

[0042] The processor 130 executes a personalized response program (hereinafter referred to as "the program") stored in the memory 120 and provides the function of controlling the hardware of the user terminal 10 by executing the program. That is, by executing the program, the processor 130 can perform necessary hardware control functions such as file system, memory allocation, network, basic libraries, timers, device control (display, media, input devices, 3D, etc.), and other utilities.

[0043] Referring to Figure 5, the processor 130 detects PII from the input user question (S110), converts the PII into neutral information (S120), and transfers the user-neutral question converted into neutral information to the management server 20 (S130).

[0044] The processor 130 can include any type of device capable of processing data. For example, it can mean a data processing device built into hardware that has physically structured circuits to perform functions expressed by code or commands contained within a program. Examples of such data processing devices built into hardware include microprocessors, central processing units (CPUs), processor cores, multiprocessors, ASICs (application-specific integrated circuits), and FPGAs (field programmable gate arrays), but the scope of the present invention is not limited to these.

[0045] The database 140 stores or provides the user terminal 10 with the necessary data, in accordance with the control of the processor 130. Exemplarily, the database 140 may include a PII database 141. Referring to Figure 6, the PII database 141 can store PII extracted from user queries. Such a database 140 may be included as a separate component from the memory 120, or it may be built in a portion of the memory 120.

[0046] Referring to Figure 6, the processor 130 may include detail modules that perform various functions in response to the execution of a personalized response program. Here, the detail modules may include a PII detection model 131 and a response generation model 132.

[0047] Referring to Figure 7, when the user terminal 10 receives update information for the management language model 210 from the management server 20, it can update the response generation model 132 or the PII detection model 131. For example, the PII detection model 131 and the response generation model 132 can perform continual learning so that the already trained models update new datasets, environments, etc., while maintaining their existing performance.

[0048] For example, the PII detection model 131 can utilize an existing language recognition model and may consist of an encoder structure. Similarly, the response generation model 132 can utilize an existing language generation model and may consist of a decoder or encoder-decoder structure. As an example, each model can apply various elements as basic components, such as transformer models and convolutional neural networks (CNNs) or recurrent neural networks (RNNs).

[0049] Referring to Figure 7, the PII detection model 131 can detect PII from the input user question, convert the PII into neutral information, and transfer the converted user-neutral question to the management server 20. At this time, the PII detection model 131 can mask the PII extracted from the user question or convert it into predetermined words to filter out the PII and generate user-neutral questions.

[0050] As an example, when the PII detection model 131 converts PII into neutral information, it can specify a representative neutral word for each type of individual classified as PII from the user question and convert the PII into that text. For example, the question, "My name is Yoo Sang-won. What is your name?" can be converted into the neutral question, "My name is Hong Gil-dong. What is your name?". As another example, when the PII detection model 131 converts PII into neutral information, the PII extracted from the user question can be converted into neutral information. <mask>Queries can be transformed by simple tokenization. For example, the question "My name is Yoo Sang-won. What is your name?" can be transformed into "My name is <mask>This can be rephrased as a neutral question: "What is your name?"

[0051] In this way, the management server 20 can learn the management language model 210 using the PII-filtered user-neutral question data and provide a common response for each neutral question pattern.

[0052] As another example, the management server 20 can learn each user's PII pattern while learning multiple user-neutral question patterns. Here, the PII pattern is, as mentioned above, a neutral word or in the sentence structure. <mask>The token's position can be used to learn for each neutral question pattern. In one embodiment, the management server 20 can provide the second user terminal's PII pattern to the first user terminal as update information. For example, in a first user question "ABC", the first user usually includes the PII at position A, but the second user includes the PII at position C. Therefore, the management server 20 can provide the first user terminal with update information indicating that there is a possibility of including the PII at both positions A and C in a question with that structure. Subsequently, the first user terminal can input a first user question similar to the second user terminal's PII pattern, rather than using its original method. At this time, the updated first user terminal's PII detection model 131 can easily recognize the PII from the first user question. This allows each user terminal 10 to improve the PII filtering function of the PII detection model 131. Furthermore, since the pattern containing the PII is updated rather than the PII of other users itself, the leakage of personal information can be prevented.

[0053] For example, the PII detection model 131 can determine whether an individual character or word constituting a user query is personal information based on initial PII pre-stored in the PII database 141, extract PII from the user query based on named entity recognition (NER), and store it in the PII database 141. For example, the process of constructing the PII database 141 must precede this, and the PII database 141 can be constructed by extracting some of the individuals classified as common PII from the user query as initial PII through a text recognition methodology such as named entity recognition (NER), or by the user directly registering initial PII. Furthermore, the PII detection model 131 can expand the PII database 141 by determining whether each individual constituting the text of a newly input user query is PII, based on the PII database 141, thereby improving the accuracy of PII detection.

[0054] For example, the PII detection model 131 can learn using a continuous learning method based on the PII database 141, which is updated by user questions and usage behavior, and can grasp PII patterns. Therefore, the PII detection model 131 can determine whether a user question contains PII and transfer user-neutral questions from which PII has been removed to the management server 20. That is, by using the PII database 141 as training data for the PII detection model 131, the PII detection model 131 can also detect text individuals associated with the PII database 141, thereby enhancing the generalization capability of PII detection. Here, the loss function applied can be a form of loss function generally used for training language encoder models. For example, the NLL (negative log likelihood) function can be applied as the loss function in the MLM (Masked language modeling) training form.

[0055] The response generation model 132 can understand the context of a user question and generate a response through natural language processing analysis. Furthermore, the response generation model 132 may consist of a language model that learns user question patterns using user questions. Here, the user question patterns may be learned for each same context based on a language distribution that includes the frequency and type of words in the user question.

[0056] For example, the response generation model 132 may use a set of user questions in natural language text form as training data. Here, the training data can be steadily updated as user question data is accumulated from user input. As a result, the response generation model 132 can learn user question patterns based on the accumulated user questions, and each user terminal 10 can provide users with personalized and natural responses.

[0057] For example, the training method for the response generation model 132 can be such that the distribution of languages ​​generated by the model in its initial state approaches the distribution of query languages ​​entered by the user. Furthermore, a loss function can be applied to the training to measure the distance between distributions so that the text distribution of the data generated by the response generation model 132 approaches the text distribution composed of user question data. Then, while continuously training with steadily updated user question data, to prevent catastrophic forgetting, an additional loss function can be applied to the training each time data is learned, which reduces the distance between the generation distribution of the model before training and the generation distribution of the model after training. For example, cross-entropy can be applied as a loss function to measure the difference between distributions.

[0058] In another embodiment, in the on-device-based personal information leakage prevention and personalized response system of the present invention, if the management server 20 and the user terminal 10 are the same entity operating the service or are collaborating, the management server 20 can perform federated learning of the management language model 210 and provide update information for each model to the user terminal 10. That is, the management server 20 can exchange learning information with each user terminal 10 and interact with them.

[0059] In an additional embodiment, if the management server 20 and the user terminal 10 are operated by different entities and do not have mutual access rights, the user terminal 10 can only forward neutral questions to the management server 20 and receive a common response to those neutral questions. That is, each user terminal 10 can use user question data to provide only personalized response services using the learned response generation model 132. In this case, the management server 20, as a third party, can only perform the role of additionally providing common responses when the user terminal 10 generates responses through the management language model 210.

[0060] In the following, explanations of the same configurations shown in Figures 1 through 7 above will be omitted.

[0061] Figure 8 is a flowchart illustrating an on-device-based personal information leakage prevention and personal response method performed by a personal response system according to another embodiment of the present invention.

[0062] Referring to Figure 8, the on-device-based prevention of personal information leakage and the personalized response method performed by the personalized response system 1 include the steps of: (S210) the management server 20 receiving user-neutral questions from multiple user terminals 10 in which PII (Personal Identifiable Information) contained in the user questions has been converted into neutral information; and (S220) the management server 20 training a management language model 210 that receives user-neutral questions and generates a common response pattern for each neutral question pattern.

[0063] Here, the management language model 210 may be configured to learn PII patterns for each neutral question pattern by utilizing the location of neutral information in the sentence structure for user-neutral questions.

[0064] A personalized response method according to one embodiment of the present invention may also be embodied in the form of a recording medium containing computer-executable commands, such as program modules executed by a computer. The computer-readable medium may be any available medium accessible by a computer, and includes all volatile and non-volatile media, removable and non-removable media. The computer-readable medium may also include computer storage media. Computer storage media include all volatile and non-volatile, removable and non-removable media embodied by any method or technique for storing information such as computer-readable commands, data structures, program modules, or other data.

[0065] Although the apparatus and methods of the present invention have been described in relation to specific embodiments, some or all of their components or operations can be embodied using a computer system having a general-purpose hardware architecture.

[0066] The above description of the present invention is illustrative, and a person with ordinary skill in the art to which the invention pertains will understand that it can be easily modified into other specific forms without altering the technical idea or essential features of the invention. Therefore, the embodiments described above should be understood to be illustrative and not limiting in all respects. For example, each component described as a single type can be implemented in a distributed manner, and similarly, components described in a distributed manner can be implemented in a combined manner.

[0067] The scope of the present invention is defined more by the claims described below than by the detailed description above, and all modifications or alterations derived from the meaning and scope of the claims and the concept of equivalents thereto should be interpreted as being included within the scope of the present invention.< / mask> < / mask> < / mask> < / mask> < / mask>

Claims

1. In preventing on-device-based personal data leaks and in personalized response systems, Personal Identifiable Information (PII) is detected from the user's questions entered. Multiple user terminals that transfer user-neutral questions obtained by converting the aforementioned PII into neutral information, and The system includes a management server that receives the user-neutral questions and trains a management language model that generates a common response pattern for each neutral question pattern, The aforementioned user terminal is A response generation model that understands the context of the user's question and generates a response through natural language processing analysis, and The PII detection model includes a method for generating user-neutral questions by masking or converting the PII extracted from the user questions into predetermined words. The response generation model is a language model that learns user question patterns using the user questions, A personalized response system in which the user question patterns are learned for each same context based on a language distribution including the frequency and type of words in the user questions.

2. The aforementioned PII detection model is Based on the initial PII pre-stored in the PII database, it is determined whether the individual characters or words constituting the user question constitute personal information, The personalized response system according to claim 1, wherein the PII is extracted from the user question based on named entity recognition (NER) and stored in the PII database.

3. In an on-device-based personal information leakage prevention and personal response system, Personal Identifiable Information (PII) is detected from the user's questions entered. Multiple user terminals that transfer user-neutral questions obtained by converting the aforementioned PII into neutral information, and The system includes a management server that receives the user-neutral questions and trains a management language model that generates a common response pattern for each neutral question pattern, The management language model learns a PII pattern for each neutral question pattern by utilizing the position of the neutral information in the sentence structure for the user-neutral question, and is a personalized response system.

4. The personalized response system according to claim 1 or 3, wherein the PII is divided into direct identifiers and quasi-identifiers that can identify a specific individual.

5. On-device-based prevention of personal information leaks and personalized responses on user terminals, Communication module, The memory in which the personalized response program is stored, and Includes a processor that executes the personalized response program, The personalized response program detects PII from the input user question, converts the PII into neutral information, and transfers the user-neutral question converted into neutral information to the management server. The aforementioned personalized response program is A response generation model that understands the context of the user's question and generates a response through natural language processing analysis, and This includes a PII detection model that generates user-neutral questions by masking or converting the PII extracted from the user questions into predetermined words. The response generation model is a language model that learns user question patterns using the user questions, The user terminal is one in which the user question patterns are learned for each same context based on a language distribution that includes the frequency and type of words in the user questions.

6. The aforementioned PII detection model is Based on the initial PII pre-stored in the PII database, it is determined whether the individual characters or words constituting the user question constitute personal information, The user terminal according to claim 5, which extracts the PII from the user question based on named entity recognition (NER) and stores it in the PII database.

7. In preventing on-device-based personal information leaks and in personal response methods performed by personal response systems, (a) The management server receives user-neutral questions from multiple user terminals in which PII (Personal Identifiable Information) included in the user questions has been converted into neutral information, and (b) The process includes the step of having the management server receive the user-neutral questions and train a management language model that generates a common response pattern for each neutral question pattern, The management language model learns a PII pattern for each neutral question pattern by utilizing the position of the neutral information in the sentence structure for the user-neutral question, and is a personalized response method.

Images