Method and System for Conversation Flow Graph-based Multi-party Group Chat Context-Aware Translation and Providing Content

Abstract

The present invention relates to a method for context-aware translation and content provision in a multi-party group chat environment performed by a processor of a system, comprising: (a) a step in which the processor collects language context information including at least one of system setting information of a user terminal, communication protocol header information, application setting information, and recipient profile information; (b) a step in which the processor determines a target language for translation by applying a pre-set weight or priority rule to the collected language context information; (c) a step in which the processor creates a volatile context window in memory including a currently input message and past utterance data within a pre-set number or time range prior to the message, and inputs this into a context-aware translation model to generate a translation result; (d) a step in which the processor extracts topic and sentiment attributes from the translation result and text data within the volatile context window; and (e) a step in which the processor determines, in real time, generative supplementary content including at least one of text, image, and video based on the extracted topic and sentiment attributes. and (f) the processor combines integrity verification data regarding the source and generation method of the content with the determined generative supplementary content, and controls and outputs the exposure method or personalization of the generative supplementary content based on a personal information processing consent signal received from a user terminal, wherein the volatile context window is not written to permanent storage and is extinguished from memory according to a preset time-to-live or session end event, and if the volatile context window is for a group chat room including multiple participants, the step (c) comprises the processor generating a speaker profile for each of the multiple participants in the group chat room;The method may be characterized by including: a step in which the processor constructs a conversation flow graph that identifies which participant uttered to which participant within the volatile context window; and a step in which the processor analyzes the conversation flow graph to selectively filter only the context related to the message to be translated, and dynamically reconstructs the volatile context window based on the filtering result.

Description

Technology Field

[0001] The present invention relates to a technology for supporting multilingual communication in a network environment, and more specifically, to an artificial intelligence-based data processing method and system that improves translation accuracy by analyzing contextual information including user terminal environment information and communication history, and generates and provides customized content associated with the context. Background Technology

[0002] As the global communication environment of modern society develops rapidly, the need for real-time translation technology to overcome language barriers is continuously increasing. However, conventional translation technologies have mostly focused on translating fragmentary sentences, resulting in technical limitations such as the inability to grasp the overall flow of a conversation or accurately identify the specific objects referred to by anaphora.

[0003] Furthermore, in the process of determining a user's language settings, conventional technology tends to rely solely on fixed settings of the terminal or web browser. Consequently, there is a problem in that the system fails to effectively reflect the fluid characteristics of multilingual users, whose preferred language changes depending on the situation.

[0004] Furthermore, content delivery systems functioning as revenue models for free services, particularly advertising systems, frequently degrade the user experience by unilaterally providing information regardless of the content or context of the conversation. In addition, technical challenges remain to be addressed, such as a lack of transparency in the data processing process to meet personal information protection regulations or the opaque use of user data.

[0005] As prior art, U.S. patent application US 2025 / 0173524 A1 proposes a method that utilizes a variable-length text window containing contextual information to improve translation quality and uses a generative model to post-edit translation results. The problem to be solved

[0006] The present invention aims to provide a method for determining the optimal target language by comprehensively analyzing various language setting signals within a user terminal in order to solve the aforementioned problems.

[0008] In addition, the present invention aims to provide a system that can maximize translation quality while protecting user privacy by utilizing a volatile memory-based context window rather than a permanent storage.

[0009] Furthermore, we aim to present an information provision and monetization model that does not compromise the user experience by providing generative content that aligns with the conversational context in a timely manner. means of solving the problem

[0010] An artificial intelligence-based context-aware translation and content provision method performed by a computing device according to an embodiment of the present invention for solving the aforementioned problem comprises: a step of collecting language context information including at least one of system setting information of a user terminal, communication protocol header information, application setting information, and recipient profile information; a step of determining a target language for translation by applying a preset weight or priority rule to the collected language context information; a step of creating a volatile context window in memory including a currently input message and past utterance data within a preset number or time range prior to the message, and inputting it into a context-aware translation model to generate a translation result; a step of extracting topic and sentiment attributes from the translation result and text data within the volatile context window; and a step of generating or searching for and determining generative supplementary content including at least one of text, image, and video in real time based on the extracted topic and sentiment attributes. The method includes the step of combining integrity verification data regarding the source and generation method of the content with the determined generative supplementary content, and controlling and outputting the exposure method or personalization of the generative supplementary content based on a personal information processing consent signal received from a user terminal, wherein the volatile context window is not written to permanent storage and is destroyed from memory according to a preset time-to-live or session termination event.

[0011] In addition, in one embodiment, the context-aware translation model of the step of generating the translation result may be a neural network model including an attention mechanism that calculates the association between words within an input sequence, and may be characterized by being trained to maintain consistency of the translation by identifying the antecedent referred to by the pronoun or demonstrator within the volatile context window.

[0012] In addition, in one embodiment, the step of determining the target language for translation may be characterized by determining the target language for translation by assigning priority in the order of explicit setting information within the application, the recipient's primary language information, and the terminal's system language information, and correcting for any conflicts by reflecting the user's recent translation selection history.

[0013] Additionally, in one embodiment, the generative supplementary content may be provided in an inline form inserted between an adjacent area of ​​the area where the translation result is displayed or between conversation bubbles so as not to disrupt the flow of conversation, and may be characterized by being selectively exposed when a transition of the topic is detected or when the response waiting time exceeds a preset threshold.

[0014] Additionally, in one embodiment, the integrity verification data includes identification information of the artificial intelligence model that generated the content, the time of generation, and an encrypted signature value for preventing tampering, and the outputting step may be characterized by providing a visual indicator on the user interface that the content was generated by artificial intelligence.

[0015] Additionally, in one embodiment, the outputting step may be characterized by providing content using a de-identified contextual matching method by excluding the user identifier and using only the contextual information within the volatile context window when the personal information processing consent signal indicates opt-out or minimal collection.

[0016] Additionally, the AI-based context-aware translation and content delivery system may be characterized by including: a language decision module that collects language-related signals from multiple sources to determine a target language; a translation engine that manages contextual data in volatile memory and performs context-aware translation; a content creation module that analyzes keywords and sentiment information from translated data and generates multimodal content corresponding thereto; and a policy management module that controls the method of providing content according to data source authentication and user personal information consent policies. Effects of the invention

[0017] According to the present invention, by applying a general-purpose language detection logic that is not dependent on a specific operating system (OS) or web browser environment, it is possible to increase user convenience in various communication environments.

[0018] In addition, by configuring to process contextual information in volatile memory rather than in permanent storage, it is possible to provide high-quality translation services that reflect context while technically complying with enhanced data security regulations.

[0019] Furthermore, by utilizing generative artificial intelligence (AI) to provide customized content that naturally blends with the conversation flow, it is possible to create added value without compromising the usability of the service. Brief explanation of the drawing

[0020] FIG. 1 is a block diagram schematically showing the overall network configuration of an artificial intelligence-based context-aware translation and content provision system according to one embodiment of the present invention. FIG. 2 is a diagram showing the interaction of a language decision module, a translation engine, a content generation module, and a policy management module, which are internal components of a computing device according to one embodiment of the present invention. FIG. 3 is a flowchart showing the overall flow of an artificial intelligence-based context-aware translation and content provision method according to one embodiment of the present invention. FIG. 4 is a conceptual diagram illustrating the creation, data processing, and destruction processes of a Volatile Context Window according to an embodiment of the present invention. FIG. 5 is a flowchart specifically illustrating the process of determining a target language for translation by applying weight and priority rules from various language context information according to an embodiment of the present invention. FIG. 6 is an example of a user interface (UI) in which generative supplementary content is provided in an inline form within a chat room along with translation results according to one embodiment of the present invention. Specific details for implementing the invention

[0021] Various embodiments are described with reference to the drawings. In the present invention, various descriptions are provided to facilitate an understanding of the invention. However, it is evident that these embodiments can be practiced without such specific descriptions.

[0022] The term "or" is intended to mean an implicit "or" rather than an exclusive "or." That is, unless otherwise specified or evident from the context, "X uses A or B" is intended to mean one of the natural implicit substitutions. In other words, if X uses A; if X uses B; or if X uses both A and B, "X uses A or B" may apply to any of these cases. Furthermore, the term "and / or" as used in this invention should be understood to refer to and include all possible combinations of one or more of the enumerated related items.

[0023] Furthermore, the terms “comprising” and / or “comprising” should be understood to mean that such features and / or components are present. However, the terms “comprising” and / or “comprising” should be understood not to exclude the presence or addition of one or more other features, components and / or groups thereof. Additionally, unless otherwise specified or clearly evident from the context to indicate a singular form, the singular in the present invention and claims should generally be interpreted to mean “one or more.”

[0024] And, the term "at least one of A or B" should be interpreted to mean "in the case of including only A," "in the case of including only B," or "in the case of being combined with the composition of A and B."

[0026] FIG. 1 is a block diagram schematically illustrating the overall network configuration of an artificial intelligence-based context-aware translation and content provision system according to an embodiment of the present invention. Hereinafter, the functions and organic connection relationships of the system (10) and each component of the present invention will be described in detail with reference to FIG. 1.

[0027] An artificial intelligence-based context-aware translation and content provision system (10) according to one embodiment of the present invention may be largely configured to include a processor (100), a user terminal (200), a receiver terminal (300), and a communication network connecting them. The system (10) supports real-time communication between multiple devices connected via the network and implements an integrated platform that goes beyond simple text translation to analyze the context of communication using artificial intelligence, and generates and provides translation results and additional content suitable for the analysis.

[0028] First, the user terminal (200) is the entity that inputs and transmits messages by executing the service of the present invention, namely a messenger application, a social network service (SNS), or a web-based chat service. The user terminal (200) can be implemented as various computing devices equipped with wired and wireless communication functions, such as a smartphone, a tablet PC, a desktop computer, a laptop, or a wearable device. The user terminal (200) can transmit not only text or voice messages input by the user but also language context information to the processor (100) via the communication network. Here, the language context information includes metadata to improve translation accuracy in addition to simple input text. Specifically, the language context information may include the operating system (OS) language settings of the user terminal (200), keyboard input settings, setting information of the currently running application, HTTP header information of the web browser (e.g., Accept-Language), and the user's past translation selection history. This information can be used as a basis for the processor (100) to determine the optimal source language and target language that are suitable for the user's intention and situation.

[0029] The recipient terminal (300) is a terminal of the counterparty that receives a message sent from the user terminal (200). The recipient terminal (300) is also connected to the communication network and can transmit its recipient profile information to the processor (100) or the communication network. The recipient profile information may include the language primarily used by the recipient, nationality, place of residence, or preferred language set within the application. In the system (10) of the present invention, the information of the recipient terminal (300) plays a very important role. This is because, unlike conventional technology which determined the translation language based solely on the sender's settings, the present invention determines the direction of translation by considering the recipient profile information together. For example, if the recipient resides in Korea but has their primary language set to English, the processor (100) can detect this and control the provision of a message translated into English rather than Korean.

[0030] The communication network above refers to a communication network that relays data transmission and reception between the processor (100), the user terminal (200), and the receiver terminal (300). This may include a mobile communication network such as 5G, LTE, LTE-A, a wireless internet such as Wi-Fi, or a wired internet such as Ethernet, and may transmit and receive data in packet units based on the TCP / IP protocol. As illustrated in FIG. 1, the communication network mediates the flow of data such as 'message and language context information', 'receiver profile information', 'translation and additional content', and 'translated message'.

[0031] The processor (100) may be a server or cloud computing device that performs the core functions of the present invention, may be composed of a plurality of physically distributed servers, or may be implemented in the form of an on-device AI processor installed within a user terminal (200). The processor (100) performs a series of processes to analyze and process data received through the communication network.

[0032] Looking at the specific operation of the processor (100), first, various language context information collected from the user terminal (200) and the receiver terminal (300) is comprehensively analyzed. The processor (100) automatically determines the language to be translated by applying a pre-set weighting logic. In this process, if app settings, system settings, receiver information, etc. conflict, the optimal language is determined according to priority rules to minimize mistranslation.

[0033] Next, the processor (100) performs context-aware translation. The processor (100) creates a volatile context window in memory that includes the current message received and past utterance data previously transmitted and received. This volatile context window is used to identify the flow of conversation, context, and the target (antecedent) indicated by the pronoun. Importantly, this context data is not written to permanent storage but is processed only temporarily in volatile memory (RAM) and then destroyed. Through this, the processor (100) can generate high-quality context translation results without permanently leaving the user's sensitive conversation content on the server.

[0034] Additionally, the processor (100) analyzes topic and sentiment attributes from text data extracted during the translation process. For example, if the conversation content is about 'travel' and the sentiment is analyzed as 'joy', the processor (100) runs a generative AI model to generate or search for additional content, such as images, emoticons, or travel destination recommendation information associated with it, in real time.

[0035] The 'translation and additional content' generated by the processor (100) is transmitted again to the user terminal (200) through the communication network. As indicated by the arrow in FIG. 1, the user terminal (200) performs 'translation result and content display' based on the received data. At this time, the generated additional content may be provided in an inline form that does not interfere with the flow of conversation on the user interface (UI), or may be displayed in a personalized form depending on whether the user consents (signal of consent to personal information processing). At the same time, the finally translated message is transmitted to the recipient terminal (300), enabling smooth communication without language barriers.

[0036] Consequently, the system (10) illustrated in FIG. 1 has the structure of an intelligent platform that goes beyond simply converting language, understands the context of communication by comprehensively considering the terminal's environment information and the user's profile, and generates and provides content based on this. This network configuration forms a structure in which the collection, analysis, processing, and distribution of data circulate organically, thereby enabling the high-precision translation and customized content provision services intended by the present invention.

[0038] FIG. 2 illustrates the data processing flow and time-series interaction between internal system modules according to an embodiment of the present invention. The artificial intelligence-based context-aware translation and content provision method according to an embodiment of the present invention can be performed by organically linking a plurality of modules as shown in FIG. 2.

[0039] First, in (a) the language context information collection step, the user terminal can collect various contextual information for language determination simultaneously with message input and transmit it to the language determination module. The information collected at this time is not limited to the input text but may include the terminal's system language settings, language settings within the application, communication protocol header information, etc. Specifically, the protocol header information may include the 'Accept-Language' field of the HTTP header or 'User-Agent' information, thereby enabling the user's actual language environment to be identified from various angles.

[0040] Next, in step (b) the target language determination step, the language determination module may apply pre-set weights and priority rules to the received language context information. For example, the language determination module may prioritize explicit setting information within the application, but if such information is absent or conflicts occur, it may determine the target language by referring to system settings or the recipient's profile information. The target language information and the original message determined through this process may be transmitted to the translation engine.

[0041] The subsequent (c) context-aware translation step can be performed by a translation engine, which is a core component of the present invention. When the translation engine receives a message, it can create or update a volatile context window on volatile memory (RAM). The volatile context window may include not only the currently input message but also past utterance data within a preset number or time range. By inputting this context window into a context-aware translation model, the translation engine can accurately identify the antecedent referred to by a pronoun or demonstrator and generate a translation result that maintains contextual consistency. In particular, since the context data is not written to a permanent storage but resides only temporarily in memory, data security can be enhanced.

[0042] After that, (d) in the topic and sentiment extraction step, the translation engine can transmit the generated translation results and contextual data to a content generation module. The content generation module can analyze the received text data to extract the core topic of the conversation and the sentiment attributes of the user. This can be performed using natural language processing (NLP) technology and can even identify the emotional nuances embedded in the context of the conversation.

[0043] (e) In the additional content generation step, the content generation module may generate or search for and determine in real time generative additional content including at least one of text, images, and videos based on the extracted topics and sentiment attributes. For example, it may generate a customized image that matches the atmosphere of the conversation or synthesize and provide informational text related to the topic of conversation.

[0044] The generated content can be finally provided to the user after undergoing (f) a verification and output control step. This step is performed by a policy management module and can largely consist of content integrity verification and personal information consent confirmation processes. The policy management module can combine integrity verification data, including the identification information of the artificial intelligence model, the time of creation, and the anti-tampering signature value, with the generated content. Additionally, by verifying a personal information processing consent signal received from the user terminal, the system can control the provision of only non-identified content using a context-matching method that excludes identifiers if the user opts out of the provision of personalized content.

[0045] Finally, the content and control signals approved by the policy management module can be transmitted in parallel to the user terminal and the recipient terminal. On the user terminal, generated supplementary content is displayed inline along with the translation, providing rich information without disrupting the flow of conversation. At the same time, the translated message is transmitted to the recipient terminal to support smooth communication.

[0046] Finally, as illustrated at the bottom of FIG. 2, when an event occurs where a conversation session ends or a preset Time-to-Length (TTL) expires, the translation engine can immediately destroy the volatile context window in memory. This ensures that the user's sensitive conversation history does not remain on the server or storage, thereby technically complying with privacy regulations and maximizing security.

[0047] In one embodiment, the AI-based context-aware translation and content provision method according to the present invention may be performed by organically linking a plurality of modules.

[0048] First, in the language context information collection stage, the user terminal can collect various contextual information for language determination simultaneously with message input and transmit it to the language determination module. The information collected at this time is not limited to simply the input text but may include at least one of the terminal's system configuration information, configuration information within the application, communication protocol header information, and recipient profile information. Specifically, the communication protocol header information may include the language setting field of an HTTP header or User-Agent information, thereby making it possible to identify the user's actual language environment from various angles.

[0049] Next, in the translation target language determination step, the language determination module may determine the translation target language by applying pre-set weights or priority rules to the received language context information. For example, the language determination module determines the target language by assigning priorities in the order of explicit setting information within the application, the recipient's primary language information, and the terminal's system language information; however, if a conflict occurs between the information, it may correct it by reflecting the user's recent translation selection history.

[0050] The subsequent context-aware translation step can be performed by a translation engine, and upon receiving a message, the translation engine can create or update a volatile context window in volatile memory. The volatile context window may include not only the currently input message but also past utterance data within a preset number or time range. By inputting this context window into a context-aware translation model, the translation engine can identify the antecedents referred to by pronouns or demonstrators and generate translation results that maintain contextual consistency. In particular, data security can be enhanced as the volatile context window is managed only in memory and is not written to permanent storage.

[0051] Subsequently, in the topic and sentiment extraction stage, the translation engine can transmit the generated translation result and text data within the volatile context window to a content generation module, and the content generation module can analyze this to extract the topic and sentiment attributes of the conversation. This can be used as basic data for subsequent content generation by identifying even the emotional nuances embedded in the context of the conversation.

[0052] In the supplementary content generation step, the content generation module may generate or search for and determine generative supplementary content in real time, including at least one of text, images, and videos, based on extracted topic and emotion attributes. At this time, the generated content may be synthesized into an advertising message or emoticon suitable for the conversational context through a large-scale language model (LLM) or an image generation model.

[0053] The generated content can be finally provided to the user after undergoing verification and output control steps. This step is performed by a policy management module, which can combine integrity verification data—including the source of the content, the time of creation, artificial intelligence model identification information, and an encrypted signature value for preventing tampering—with the determined generated supplementary content. Additionally, the exposure method or whether the generated supplementary content is personalized can be controlled based on a personal information processing consent signal received from the user terminal. If the personal information processing consent signal indicates an opt-out for personalization, the user identifier can be excluded, and content using a de-identified context-matching method can be provided using only context information within the volatile context window.

[0054] Finally, the content and control signals approved by the policy management module can be transmitted to the user terminal. In this case, the generated supplementary content may be provided in an inline form, inserted between speech bubbles or in an area adjacent to where the translation result is displayed, so as not to disrupt the flow of conversation. Furthermore, when an event occurs in which the conversation session ends or a preset Time-To-Live expires, the volatile context window is not written to permanent storage but is immediately destroyed from memory, thereby protecting user privacy and complying with data security regulations.

[0056] FIG. 3 is a flowchart illustrating the overall flow of an artificial intelligence-based context-aware translation and content provision method according to an embodiment of the present invention. Hereinafter, with reference to FIG. 3, the specific operation and technical features of each step constituting the method will be described in detail. The method according to an embodiment of the present invention can be performed chronologically by a processor of a computing device.

[0057] First, in step (a), the processor can collect language context information containing information from various sources from the user terminal. This language context information is not limited to simple input text but can encompass environmental variables that determine the accuracy of the translation. Specifically, the processor can collect system configuration information of the user terminal (OS language, etc.), communication protocol header information (HTTP User-Agent, Accept-Language, etc.), explicit configuration information within the application, and profile information of the recipient who is the communication partner. This collection of complex information is intended to overcome the limitations of conventional technology that relied on a single configuration value and to comprehensively understand the user's actual language environment.

[0058] Next, in step (b), the processor can determine the target language by applying pre-set weights or priority rules to the collected language context information. In this process, the processor can perform logic to resolve conflicts between multiple language determination factors. For example, the target language can be determined by assigning priorities in the order of application setting information, the recipient's primary language information, and the terminal's system language information, and if conflicts occur between the information, the conflict can be corrected by reflecting the user's recent translation selection history.

[0059] Step (c) is a step in which context-aware translation, which is the core of the present invention, is performed. The processor may create a Volatile Context Window in memory that includes not only the currently input message but also past utterance data within a preset number or time range prior to the message. The Volatile Context Window is managed only on volatile memory (RAM) rather than permanent storage (HDD / SSD), thereby ensuring data security. The processor may input this window data into a Context-Aware Translation Model to generate a translation result. At this time, the Context-Aware Translation Model may be a neural network model that includes an attention mechanism for calculating the association between words within an input sequence, and in particular, may be a model trained to maintain contextual consistency of the translated word by identifying the antecedents referred to by pronouns or demonstrators within the Volatile Context Window.

[0060] In the subsequent step (d), the processor can extract topic and sentiment attributes from the generated translation result and the text data within the volatile context window. This is a process for understanding the context of a conversation beyond simple translation, and through natural language processing (NLP) technology, it is possible to precisely analyze whether the conversation is about 'travel' or whether the user's emotion is 'joy'.

[0061] In step (e), the processor may generate or search for and determine Generative Additional Content (GNP) in real time, including at least one of text, images, or videos, based on the extracted topic and sentiment attributes. The processor may utilize a Large Language Model (LLM) or an image generation model to synthesize advertising messages or customized emoticons suitable for the conversation context in real time. Additionally, the Generative Additional Content may be provided in an inline form inserted between adjacent areas of translation results or between speech bubbles so as not to impair the user experience, and may be controlled to be selectively exposed when a topic switch is detected or when the response waiting time exceeds a threshold.

[0062] Finally, in step (f), the processor can perform verification and output control for the determined generated additional content. The processor can combine integrity verification data, including the source and generation method of the content, the time of generation, and an encrypted signature value for preventing tampering, with the content. This allows for the provision of a visual indicator on the user interface that the content was generated by artificial intelligence. Additionally, the processor can control the output method based on a personal information processing consent signal received from the user terminal. If the user chooses to opt out of personalization or minimal collection, the processor can control the provision to provide only de-identified content using a contextual matching method that excludes user identifiers and utilizes only contextual information within a volatile context window.

[0063] When all of the above processes are completed or an event occurs in which the conversation session ends, the volatile context window is immediately destroyed from memory without being written to permanent storage according to a preset time-to-live lifecycle, thereby strongly protecting the user's privacy.

[0064] In one embodiment, referring to FIG. 3, the artificial intelligence-based context-aware translation and content provision method of the present invention may be composed of steps performed chronologically by a processor of a computing device.

[0065] First, in step (a), the processor can collect language context information from the user terminal, including not only simple input text but also various environmental variables that affect the accuracy of the translation. Specifically, the language context information may include system configuration information of the user terminal, communication protocol header information (HTTP User-Agent, etc.), explicit configuration information within the application, and recipient profile information, thereby enabling a three-dimensional understanding of the user's language environment.

[0066] Next, in step (b), the processor may determine the target language for translation by applying pre-set weights or priority rules to the collected language context information. At this time, the target language is determined by assigning priorities in the order of application setting information, the recipient's primary language information, and the terminal's system language information, and a logic may be executed to correct for conflicts between the information by reflecting the user's recent translation selection history.

[0067] In step (c), context-aware translation can be performed using a volatile context window created on volatile memory (RAM). The processor can generate a translation result by inputting the window data, which includes the current message and past utterance data, into a context-aware translation model. In particular, the model can maintain contextual consistency of the translated word by identifying the antecedent referred to by a pronoun or demonstrator through an attention mechanism.

[0068] In the subsequent step (d), the processor can extract topic and sentiment attributes of the conversation from the generated translation result and data within the volatile context window using natural language processing techniques.

[0069] In step (e), the processor may generate or retrieve generative supplementary content in real time, including at least one of text, images, and videos, by utilizing a large-scale language model (LLM) or an image generation model based on the extracted topic and sentiment attributes. The content may be provided inline in an area adjacent to the translation result, or may be selectively exposed at the time of topic switching.

[0070] Finally, in step (f), the processor combines integrity verification data including the source of the content and an anti-tampering signature value, and can control the output method according to the user terminal's consent signal for personal information processing. If the user selects to opt out of personalization, content using a de-identified context matching method that excludes user identifiers and utilizes only context information may be provided. Furthermore, upon termination of the conversation session, the volatile context window is not written to permanent storage but is immediately destroyed from memory to protect the user's privacy.

[0072] FIG. 4 is a conceptual diagram illustrating the creation, data processing, and destruction processes of a Volatile Context Window according to an embodiment of the present invention. Hereinafter, with reference to FIG. 4, the volatile memory-based data processing mechanism, which is a core security technology of the present invention, and its technical significance will be explained in detail.

[0073] A system according to one embodiment of the present invention prioritizes the protection of user personal information and data security. To this end, it may adopt an architecture that exclusively processes data required for translation and context analysis only on volatile memory (RAM) rather than on non-volatile storage. The area indicated by the dotted line in FIG. 4 represents a physical volatile memory area, and the context-aware translation execution process and the context window update process of the present invention may be characterized as taking place only within this volatile memory area.

[0074] First, regarding the process of creating and updating a volatile context window in detail, the system can receive the corresponding text data as a new message input occurs from a user terminal. At this time, the system of the present invention does not follow the conventional log storage method of recording the input message to a non-volatile storage medium such as a database (DB) or a hard disk, but can immediately transmit it to a context window update module allocated in memory. The context window update module can operate based on a First-In First-Out (FIFO) sliding window algorithm or a circular queue data structure.

[0075] Specifically, when a new conversation message is entered, the system can add it to the latest data area of ​​the context window. At the same time, the oldest data that exceeds the pre-configured window size—for example, the number of messages that can be stored or the time range to be maintained—can be processed by being deleted or overwritten in memory. Through this update mechanism, the system can prevent memory overflow caused by infinitely growing data and efficiently maintain only the most relevant context of the currently ongoing conversation.

[0076] Next, in the context-aware translation execution stage, the translation engine can perform translation operations by referencing the entire data sequence residing within the updated volatile context window. This can be achieved not by simply translating a single currently input sentence independently, but by inputting the previous utterances included within the window together into the context-aware translation model. Through this, the translation model can clearly identify exactly which object a pronoun or demonstrative hidden within the conversation flow refers to, and can generate a consistent translation result that reflects contextual nuances. The generated translation result and the associated supplementary content are output to the user and recipient terminal, and this data circulation process in memory can be performed repeatedly while the conversation continues.

[0077] The most important technical feature of the present invention lies in the data destruction and destruction mechanism illustrated on the right side of FIG. 4. The system can physically destroy the corresponding context window from memory as soon as a specific event, such as session termination or the expiration of the Time-To-Live (TTL), is detected. Here, session termination may include explicit termination actions such as a user leaving a chat room or closing an application, and TTL expiration may refer to the case where there is no additional input for a preset threshold time (e.g., 10 minutes) after the last message input.

[0078] When such an expiration event occurs, the system can render the data unrecoverable by releasing the reference to the context window area allocated to volatile memory (RAM) or by overwriting it with random data. Consequently, as indicated by the data permanent expiration and no storage record block in Fig. 4, the user's sensitive conversation content leaves no trace on any form of permanent storage media, such as the server's hard disk, SSD, or backup server.

[0079] This structure has the characteristic that data exists in memory only when computation is required and evaporates immediately once its purpose is achieved or its validity period expires. By ensuring compliance with personal information protection regulations at the technical architecture level, this provides a powerful security effect that fundamentally blocks the possibility of past conversation data leakage even in the event of security incidents such as hacking or server hijacking.

[0080] In one embodiment, referring to FIG. 4, the system of the present invention may adopt an architecture that exclusively processes data required for translation and context analysis only on volatile memory (RAM) rather than in permanent storage for the protection of user personal information and data security. The dotted area of ​​FIG. 4 represents a physical volatile memory area, and the execution of context-aware translation and the updating of the context window are characterized by taking place only within this area.

[0081] First, during the creation and update phase of the volatile context window, the system can immediately transmit messages input from the user terminal to a context window update module in memory without writing them to a non-volatile storage medium such as a database or hard disk. The context window update module operates based on a First-In, First-Out (FIFO) sliding window algorithm or a circular queue data structure; when a new message is input, it is added to the latest data area, and memory overflow can be prevented by deleting or overwriting the oldest data that exceeds a preset size or time range.

[0082] Next, in the context-aware translation execution stage, the translation engine can perform translation operations by referencing the entire data sequence within the updated volatile context window. This is intended to clearly identify the objects referred to by pronouns or demonstrators and to generate consistent translation results that reflect contextual nuances by inputting not only the currently input sentence but also the previous utterances within the window into the context-aware translation model.

[0083] Finally, during the data extinction and destruction phase, the system can physically expel the corresponding context window from memory as soon as specific events, such as session termination or Time-to-Live (TTL) expiration, are detected. Here, TTL expiration may include cases where there is no additional input for a preset threshold time (e.g., 10 minutes) following the last input. Upon the occurrence of an extinction event, the system can render the data unrecoverable by releasing references to the memory area or overwriting it with random data. Consequently, user conversation content is not recorded on any permanent storage medium, and since the data evaporates immediately after the computational purpose is achieved, the possibility of past data leakage can be fundamentally prevented even in the event of security incidents such as hacking.

[0085] FIG. 5 is a flowchart specifically illustrating the process of determining a target language for translation by applying weight and priority rules from various language context information according to an embodiment of the present invention. Hereinafter, with reference to FIG. 5, a logic for determining the optimal translation language by comprehensively considering the user and the communication environment without relying on fragmentary setting values ​​will be described in detail.

[0086] A processor according to one embodiment of the present invention may first perform a step of collecting language context information prior to starting a translation process. This corresponds to the top step of FIG. 5, and the processor may collect complex data including application setting information within the user terminal, system language settings, and profile information of the recipient who is the communication partner. This collection of information is a process of securing basic data to determine which language the user actually intends to communicate in, going beyond simply reading the currently set language code.

[0087] Based on the collected information, the processor can perform sequential judgments according to a pre-set priority logic. As illustrated in FIG. 5, the first judgment criterion may be the presence or absence of application settings (app settings). If a user has explicitly specified a particular language within the messenger or communication application, the processor may select this as the highest priority candidate, as this reflects the user's most direct intent. For example, even if the user's smartphone system language is set to Korean, the user may have set the global business messenger app to use English; in this case, the app setting value may be prioritized.

[0088] If no separate language setting exists within the application, the processor may check for the presence of recipient information as a second criterion. This approach takes into account the fact that the present invention is a translation technology for communication, taking into account the language environment of the conversation partner. If the recipient's profile information specifies the primary language used or if the recipient's country code is verified, the processor may select this as a candidate language. This provides the convenience of allowing the system to automatically prepare a translation tailored to the recipient's language, even if the sender does not know the recipient's language.

[0089] If both the aforementioned application settings and recipient information are absent, the processor may select the terminal's system language as the last candidate. This reflects the most common user environment, and in the absence of separate settings, the language used by the user throughout the terminal is used as the standard language for translation.

[0090] A characteristic part of the present invention is the conflict determination and history analysis process illustrated at the bottom of FIG. 5. The processor can determine whether previously selected candidate languages ​​conflict with each other or whether inconsistencies occur between various collected contextual information. For example, complex situations may occur, such as when the app settings are in English, but the recipient is Korean, and the user's past conversation patterns mainly consist of Korean.

[0091] If such a conflict is detected, the processor may perform a history analysis and correction step. In this step, the processor may analyze the user's past translation selection history or logs of language changes in recent conversations. If the user frequently manually changes the language recommended by the system, the processor may correct the final target language by assigning weights to such change history. On the other hand, if no conflict is detected, the processor may immediately determine the selected candidate language as the final target language.

[0092] Through such hierarchical and intelligent decision logic, the present invention can eliminate the inconvenience of users having to manually set the translation language every time. In addition, it can minimize mistranslations or translation errors into the wrong language that occurred by relying only on fixed settings of the OS or browser, and automatically provide an optimal translation environment suitable for the situation and context.

[0093] In one embodiment, referring to FIG. 5, the processor of the present invention can determine the optimal target language for translation by applying weight and priority rules to various language context information without relying on fragmentary setting values.

[0094] First, the processor may collect complex language context information, including application setting information, system language settings, and recipient profile information of the user terminal, prior to the initiation of the translation process. Based on the collected information, the processor may perform sequential judgments according to a pre-configured priority logic. Specifically, application setting information reflecting the user's explicit intent may be selected as the highest priority candidate, and if such settings are absent, recipient profile information may be selected as the next highest priority candidate by considering the language environment of the communication counterpart. If all of the above information is absent, the terminal's system language may be adopted as the final candidate.

[0095] Additionally, the processor may perform a conflict detection step to determine whether the selected candidate language conflicts with other contextual information. For example, a conflict may be determined if there is a discrepancy between the application's set language and the recipient's primary language, or if the user's past conversation patterns differ from the selected candidate language. If such a conflict is detected, the processor may perform a history analysis step to analyze the user's past translation selection history or recent language change logs, and correct the candidate language by assigning weights to them. On the other hand, if no conflict is detected, the processor may immediately determine the selected candidate language as the final target language, thereby automatically providing an optimal translation environment that matches the situation and context.

[0097] FIG. 6 is an example of a user interface (UI) in which generative supplementary content is provided in an inline form within a chat room along with translation results according to an embodiment of the present invention. Hereinafter, specific operations and technical implementation forms in terms of user experience (UX) will be described with reference to FIG. 6.

[0098] Referring to FIG. 6, a conversation session with the other party (James) is active on the messenger application screen of the user terminal. The system of the present invention receives the original message (Source Text) received from the other party, "Hey, are you free for Sushi tonight? I know a great place nearby." and simultaneously performs context-aware translation to generate a translation result (Target Text) "How about sushi for dinner tonight? I know a good place nearby." As illustrated in the figure, the translation result may be provided in the form of a separate speech bubble with a tag labeled "AI Translation" immediately below the original message, taking into consideration user visibility.

[0099] At this time, the content generation module of the present invention analyzes data within a volatile context window to identify that the core topic of the current conversation is 'Sushi' and 'dinner,' and can identify positive sentiment or suggestion intentions regarding this. Based on the results of this context analysis, the system can search for or generate additional generative content in real time, such as "Top 3 Popular Sushi Restaurants Nearby," which is information predicted to be needed by the user, going beyond simple translation.

[0100] The 'context-matched recommendation' area located in the center of Fig. 6 specifically illustrates the inline content provision method described in Claim 4 of the present invention. The generative additional content can be rendered in the form of a card or widget that is naturally inserted into the idle space between the previous message (translation) and the subsequent message (user response), rather than interrupting the conversation flow by causing a pop-up window or screen transition. The content can be configured in a multi-modal form that includes not only text information but also related images such as "Sushi Img," thereby enhancing information delivery.

[0101] In addition, it can be seen that a visual indicator reading "AI Generated" and a lock icon are displayed at the top right of the supplementary content area. This is an operation based on the integrity verification data described in claim 5, and is a technical device that transparently notifies the user that the content has been automatically generated by an artificial intelligence model. This explicit indication can help the user clearly recognize the source of the content and judge the reliability of the information.

[0102] Finally, the sight of a user responding, "Great! What time should we meet there?" after reviewing the content suggests that the context-linked content provided by the system of the present invention does not hinder the flow of actual conversation but rather acts as a medium that facilitates decision-making or enriches the conversation. In other words, the present invention seamlessly integrates translation and information provision functions within a single interface, thereby simultaneously resolving language barriers and enhancing user convenience.

[0103] In one embodiment, referring to FIG. 6, the system of the present invention can provide translation results and generative additional content in an inline form to maximize user experience.

[0104] First, when a user terminal receives an original message from another party, such as "Hey, are you free for Sushi tonight?", the system immediately performs context-aware translation to generate a translation result, "How about sushi for dinner tonight?", and displays this in the form of a speech bubble with an 'AI translation' tag attached at the bottom of the original message to ensure visibility. At this time, the content generation module analyzes data within the volatile context window to identify that the core topics of the conversation are 'Sushi' and 'dinner', and by grasping the intent of a positive suggestion, it can determine whether to generate or search for customized information in real time, such as "Top 3 popular sushi restaurants nearby."

[0105] The aforementioned generative supplementary content can be rendered inline in the form of cards or widgets that are naturally inserted into the space between the translated message and the user's response input window, without triggering pop-up windows or screen transitions. It can also be configured in a multimodal form combining text and images to enhance information delivery. Additionally, a visual indicator and icon labeled 'AI Generated' may be displayed in the content area; this serves as a technical mechanism to transparently notify the user that the content was generated by an artificial intelligence model based on integrity verification data. Consequently, this interface provides useful information in a timely manner to aid user decision-making without disrupting the flow of conversation, thereby simultaneously overcoming language barriers and increasing convenience.

[0107] Embodiments that can be extended from the embodiments of the present invention are disclosed.

[0108] According to another embodiment of the present invention, the system can be extended to provide context awareness functions in a multi-party chat environment as well as in a one-on-one conversation. In this case, the processor can generate an individual 'speaker profile' for each of the multiple participants in the group chat room and construct a 'conversation flow graph' that identifies who spoke to whom within a volatile context window. The processor can analyze the graph to maintain translation accuracy by selectively filtering only the relevant context, even when the conversation context between specific participants is mixed with the entire group conversation. For example, in a group chat room where situations are mixed such that A and B talk about travel and C and D talk about work, the processor can dynamically reconstruct the context window by weighting only B's previous utterance when translating A's message.

[0109] In another embodiment, the system can be extended beyond text-based communication to voice call or video conference environments. In this case, the user terminal converts real-time voice data into text (STT: Speech-to-Text) and transmits it to the processor, and the processor performs context-aware translation based on the converted text, and then synthesizes it into speech (TTS: Text-to-Speech) suitable for the recipient's language and transmits it. At this time, the user's 'sentiment' attribute analyzed by the content generation module can be reflected in the speech synthesis process. For example, if the sender's sentiment is analyzed as 'urgency,' 'prosodic translation' is performed by increasing the speech speed of the translated voice or emphasizing the tone, thereby providing a multimodal communication environment that conveys non-verbal nuances beyond simple meaning.

[0110] In an additional embodiment, the present invention may be implemented with a 'Hybrid Edge Computing' architecture to further enhance data security. In this embodiment, 'context-aware translation' and 'volatile context window management,' which are likely to contain sensitive personal information, are performed independently by an on-device AI processor within the user terminal, and only 'generative supplementary content generation (e.g., image generation, complex information retrieval),' which requires high computational power, may be delegated to a cloud server. In this case, data transmitted from the terminal to the server may be converted into a de-identified and vectorized 'Context Embedding' value form and transmitted, rather than as original text. Through this, the server can generate context-appropriate content and return it to the terminal even in a zero-knowledge state, thereby simultaneously achieving privacy protection and the provision of high-quality services. Explanation of the symbols

[0111] 10 : System 100 : Processor 200 : User terminal 300 : Recipient terminal

Claims

Claim 1 A method for context-aware translation and content provision in a multi-party group chat environment performed by a processor of a system, comprising: (a) a step in which the processor collects language context information including at least one of system setting information of a user terminal, communication protocol header information, application setting information, and recipient profile information; (b) a step in which the processor determines a target language for translation by applying a pre-set weight or priority rule to the collected language context information; (c) a step in which the processor creates a volatile context window in memory including a currently input message and past utterance data within a pre-set number or time range prior to the input message, and inputs this into a context-aware translation model to generate a translation result; (d) a step in which the processor extracts topic and sentiment attributes from the translation result and text data within the volatile context window; (e) a step in which the processor generates or searches for and determines generative supplementary content including at least one of text, image, and video in real time based on the extracted topic and sentiment attributes. and (f) the processor combines integrity verification data regarding the source and generation method of the content with the determined generative supplementary content, and controls and outputs the exposure method or personalization of the generative supplementary content based on a personal information processing consent signal received from a user terminal; wherein the volatile context window is not written to permanent storage and is extinguished from memory according to a preset time-to-live or session end event, and if the volatile context window is for a group chat room including multiple participants, the step (c) comprises the processor generating a speaker profile for each of the multiple participants in the group chat room;A method for context-aware translation and content provision in a multi-party group chat environment, characterized by comprising: a step in which the processor constructs a conversation flow graph that identifies which participant speaks to which participant within the volatile context window; and a step in which the processor analyzes the conversation flow graph to selectively filter only the context related to the message to be translated, and dynamically reconstructs the volatile context window based on the filtering result.

Images