Data processing system

Abstract

The present application relates to a virtual office system for realizing global collaboration of cross-national teams, which provides an efficient and smooth global collaboration environment beyond physical, language, and cultural barriers by adopting the following configuration: a virtual space generation unit generates a meta-universe space connecting multiple sites based on attribute information of users such as language and cultural background, a communication support unit performs real-time translation, cultural adaptation, and emotion recognition using generative AI, and a project progress management unit analyzes activity history and past data to autonomously propose progress optimization and cultural friction mitigation strategies.

Description

Technical Field

[0001] The technology disclosed herein relates to a system. Background Technology

[0002] Patent Document 1 discloses a personalized chatbot control method executed by at least one processor, the method comprising: receiving user speech; adding the user speech to a prompt containing instructions related to a chatbot role; encoding the prompt; and inputting the encoded prompt into a language model to generate chatbot speech in response to the user speech.

[0003] Patent document 1: Japanese Patent Application Publication No. 2022-180282.

[0004] The problem this invention aims to solve is the communication barriers faced by enterprises and multinational teams operating globally, corporations and individual business owners promoting remote work, and international freelance communities. These barriers stem from friction caused by language barriers and cultural differences, leading to inefficiencies in collaboration due to time zone and geographical limitations, high costs associated with acquiring global talent (relocation expenses and visa procedures, etc.), and a lack of speed and flexibility in executing international projects. Communication tools used as alternatives to the status quo often lack sufficient real-time translation and cultural adaptation capabilities, thus failing to adequately eliminate language and cultural barriers and hindering the effective utilization of global talent and efficient project execution by multinational teams.

[0005] Furthermore, in the existing remote work environment, the inability to completely overcome physical mobility and geographical limitations also limits the promotion of international talent mobility and new business development. Therefore, the technical problem to be solved by this invention is to comprehensively and fundamentally solve these problems and provide a new virtual office environment that transcends physical limitations and language and cultural barriers, enabling multinational teams to collaborate smoothly and efficiently. Summary of the Invention

[0006] To address the aforementioned challenges, a system for enabling global collaboration among multinational teams is provided. This system comprises a virtual space generation department, a communications support department, and a project schedule management department. The virtual space generation department accepts access from multiple physically separated locations or user terminals, registers language and cultural background information as attributes for each user, and generates virtual office spaces that the multinational team can participate in simultaneously based on this information. The communications support department uses generative AI to translate text or voice messages between users in real time, automatically adjusting expressions and subtle cultural nuances based on each user's cultural background. Furthermore, it applies generative AI-based emotion recognition algorithms to analyze the emotional state of interactive messages, visualizing the quality and atmosphere of communication. The project progress management department maintains users' activity history and past project data in the virtual space as long-term memory. Through high-precision data analysis based on generative AI, it monitors progress, extracts issues, optimizes task allocation, detects signs of cultural friction and proposes mitigation strategies. Furthermore, by having generative AI comprehensively and autonomously execute all these processes, it provides a new virtual office environment that transcends physical limitations and language and cultural barriers, enabling multinational teams to collaborate smoothly and efficiently. Attached Figure Description

[0007] Figure 1 This is a conceptual diagram illustrating an example of the configuration of a data processing system according to the first embodiment.

[0008] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0009] Figure 3 This is a conceptual diagram illustrating an example of the data processing system configuration in the second embodiment.

[0010] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0011] Figure 5 This is a conceptual diagram illustrating an example of the data processing system configuration in the third embodiment.

[0012] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0013] Figure 7 This is a conceptual diagram illustrating an example of the data processing system configuration in the fourth embodiment.

[0014] Figure 8This is a conceptual diagram illustrating an example of the functions of the main parts of the data processing device and robot according to the fourth embodiment.

[0015] Figure 9 It represents an emotion graph that maps multiple emotions.

[0016] Figure 10 It represents an emotion graph that maps multiple emotions.

[0017] Explanation of reference numerals in the attached figures Data processing systems 10, 210, 310, and 410 12 Data processing device 14 Smart devices 214 Smart Glasses 314 Head-mounted terminal 414 Robot. Detailed Implementation

[0018] Hereinafter, an example of an implementation of the system involved in this disclosure will be described with reference to the accompanying drawings.

[0019] First, let's explain the terms used in the following description.

[0020] In the following embodiments, the processor (hereinafter referred to as "processor") can be a single computing device or a combination of multiple computing devices. Furthermore, a processor can be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0021] In the following implementation, the labeled RAM (Random Access Memory) is a memory that temporarily stores information and is used by the processor as working memory.

[0022] In the following embodiments, the labeled memory is one or more non-volatile storage devices used to store various programs and various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disk (e.g., hard disk) or magnetic tape, etc.

[0023] In the following implementation, the labeled Communication I / F (Interface) is an interface that includes a communication processor and an antenna, etc. The Communication I / F is responsible for communication between multiple computers. Examples of communication standards applicable to the Communication I / F include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0024] In the following implementation, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects more than three items, the same approach as "A and / or B" applies.

[0025] [First Implementation] Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0026] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. An example of the data processing device 12 is a server.

[0027] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0028] The smart device 14 includes a computer 36, a receiver 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the receiver 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0029] The receiving device 38 includes a touchscreen 38A and a microphone 38B, etc., for receiving user input. The touchscreen 38A receives user input generated by contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input generated by sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing user input received through the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0030] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0031] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0032] Figure 2 An example of the main functions of the data processing device 12 and the smart device 14 is shown.

[0033] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 based on the specific processing program 56 executed on the RAM 30.

[0034] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0035] In the smart device 14, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The receive / output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart device 14 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0036] In addition, other devices besides the data processing device 12 may also have a data generation model 58.

[0037] For example, a server device (e.g., a generation server) may have a data generation model 58.

[0038] In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with a server device that has the data generation model 58. Furthermore, the data processing device 12 can be a server device or a user-owned terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing performed by the data processing system 10 of the first embodiment will be described.

[0039] (Example 1) The specific processing flow in Example 1 will be described below. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server" and the smart device 14 is referred to as the "terminal".

[0040] (Implementation methods for carrying out the present invention) As an embodiment of the present invention, a system is constructed through the collaborative operation of a server and a terminal to create a virtual office environment that enables multinational teams to collaborate across physical limitations and language and cultural barriers. The following provides a comprehensive and detailed description of each component, the utilization method of generative AI, the role division between the server and the terminal, and specific application examples and scenarios. Specifically, this system employs a distributed computing architecture. The system utilizes a parallel computing cluster composed of multiple computing nodes. The system has an inference server that hosts a large language model. This inference server executes a Transformer architecture with a self-attention mechanism. The system receives multidimensional feature vectors sent from the terminal. The system converts the feature vectors into tensor form. The system inputs the tensor data into the large language model. The system uses, for example, 512-dimensional word embedding vectors as input data. The system uses, for example, 1024-dimensional speech feature vectors as input data. The system uses, for example, 256-dimensional user profile vectors as input data. The system combines these input vectors to generate a composite tensor. The system provides the composite tensor to the input layer of the large language model. The system performs nonlinear transformations in the hidden layers of the large language model. This system outputs a probability distribution vector from the output layer. It outputs, for example, the probability score of the next word, a classification label for sentiment state, and a coordinate matrix of spatial layout. The system decodes the highest-probability tag sequence from the probability distribution vector. This decoded tag sequence is then converted into control commands, which are sent to each terminal. This drives the rendering engine on the terminal side. The rendering engine renders the grid data of the virtual space. The rendered image data is then output to the display. This system reduces processing latency through such hardware and software coordination. It minimizes network bandwidth consumption, thus achieving real-time virtual space synchronization. Unlike previous rule-based processing, this system analyzes complex cultural contexts through vector operations in high-dimensional space. It uses the cross-entropy error function to pre-optimize the weight parameters of a large language model. It updates the weights using gradient descent. This significantly improves translation accuracy. This system is used in online conferencing systems for multinational corporations and in virtual laboratories for international collaborative research projects. This system is used as a virtual response window in global customer support centers.

[0041] First, the virtual space generation is implemented on the server side. The server receives user information sent from user terminals (name, affiliation, language, cultural background, time zone, job title, professional field, past project experience, preferred work style, virtual avatar settings, preferred interface design, etc.) and generates a virtual office space optimized for each user based on this information. For example, in the case of a company with branches in multiple regions such as Asia, Europe, North America, South America, Africa, and Oceania, the server integrates the physical location information and network information of each branch to construct a metaverse space that virtually connects the branches. Within the virtual space, various areas are set up, such as meeting rooms, private cubicles, lounges, project rooms, cafeterias, activity halls, libraries, rest areas, reception areas, whiteboard areas, and presentation rooms, which users can freely move from their terminals to. For example, dedicated virtual meeting rooms or group discussion rooms can be automatically generated for each project, and the layout and interface can be dynamically optimized based on the participants' attributes. In addition, the lighting and background music, virtual avatar clothing, and display language within the virtual space can be automatically switched according to the user's time zone and work period. Specifically, this virtual space generation unit executes on a server equipped with a dedicated graphics processing unit. This virtual space generation unit inputs text data received from the user terminal into a natural language processing model. This natural language processing model converts words into 768-dimensional dense vectors. This virtual space generation unit converts categorical variables into one-hot vectors. This virtual space generation unit combines these vectors to generate a 2048-dimensional user feature tensor. This virtual space generation unit inputs this tensor into a generator of a generative adversarial network. This generator has multiple transposed convolutional layers. This generator generates voxel data of a three-dimensional space from the input tensor. The voxel data has, for example, a resolution of 256 x 256 x 256. This virtual space generation unit converts the voxel data into a polygonal mesh. This virtual space generation unit outputs the vertex coordinate matrix of the polygonal mesh. This virtual space generation unit outputs, for example, texture mapping coordinates of walls and floors as output data. This virtual space generation unit outputs, for example, environmental parameter vectors of light sources as output data. This virtual space generation unit outputs, for example, the initial configuration coordinates of the virtual avatar as output data. This virtual space generation unit sends the vertex coordinate matrix to the rendering engine. This rendering engine calculates the depth map of virtual space based on the vertex coordinate matrix. It uses this depth map to perform occlusion culling. This virtual space generation unit significantly reduces the computational load required for rendering through this process. It also reduces server memory consumption and compresses the amount of data transmitted over the network. This minimizes rendering latency on the client side. This virtual space generation unit is used for building virtual headquarters offices for multinational corporations, for generating venues for international online exhibitions, and for building virtual campuses for global educational institutions.

[0042] The Communications Support Department primarily operates on the server side, but some user interface and real-time display processing is also executed on the terminal side. The department uses generative AI to perform real-time translation of messages between users, including text chat, voice calls, video conferencing, whiteboard sharing, file sharing, and emotional expressions via stickers or emoticons. For example, content spoken in Japanese is simultaneously translated into multiple languages, including English, Chinese, Spanish, French, German, Arabic, and Hindi, and then distributed to various user terminals. The translation processing uses a large language model based on generative AI, supporting technical terms, industry-specific expressions, abbreviations, slang, dialects, business terms, company internal terminology, and project-specific abbreviations. For instance, in IT projects, terms such as "API," "deployment," and "Agile" are translated into appropriate technical terms in each language. Furthermore, generative AI automatically adapts to different cultures by taking into account each user's cultural background, past communication history, business etiquette in different countries, religious considerations, gender expressions, the distinction between honorific and informal expressions, the interpretation of humor or sarcasm, and the meaning of nonverbal expressions (gestures or facial expressions). For example, it can convert honorific expressions from Japanese culture to informal expressions from American culture, or adjust euphemisms from European culture to direct expressions from Asian culture, or convert them to expressions reflecting religious considerations within the Islamic cultural sphere. For instance, if a member in India says "We'll discuss it," it can convey the clear intention to a member in the United States as "I'll get back to you soon." Additionally, using generative AI-based emotion recognition algorithms, it analyzes positive, negative, stressful, tense, reassured, angry, confused, expectant, joyful, disappointed, and fatigued emotional states from the tone of voice, facial expressions, gestures, intonation, speaking speed, typing speed, and the frequency of emoji or sticker usage in messages or speeches. These emotions are visualized on the terminal screen through icons, color differentiation, charts, heatmaps, and changes in the virtual avatar's facial expressions. For example, if some members experience increased stress during a meeting, the AI ​​automatically inserts a coordinating message, such as "The discussion seems intense. We suggest a 5-minute break," or "Opinions seem to differ. Let's clarify our positions," to ease the discussion or propose a break. Furthermore, the AI ​​analyzes the number of times a user speaks, the duration of their speech, the diversity of their content, and their responses to others. For members who speak less, it can automatically generate encouraging messages such as "We'd also like to hear Mr. / Ms. ○○'s opinion." Specifically, this communications support department implements a pipeline that integrates speech recognition models, large-scale language models, and convolutional neural networks for emotion recognition. This communications support department uses, for example, speech waveform data with a sampling rate of 16000 Hz as input data. This communications support department uses, for example, a two-dimensional tensor of Mel-frequency cepstral coefficients extracted from the speech waveform as input data. This communications support department uses, for example, facial feature point coordinate vectors extracted from camera images as input data.This communication support unit inputs speech waveform data into a speech recognition model to generate a text tag sequence. This unit then combines this text tag sequence with the user's cultural attribute vector. The resulting tensor is input into a large-scale language model. This large-scale language model has an encoder and decoder structure. It calculates contextual dependencies using a self-attention mechanism. The unit outputs, for example, a text tag sequence translated into the target language. It also outputs, for example, a candidate list of alternative expressions after cultural adaptation. Furthermore, it outputs, for example, a probability distribution vector representing positive or negative emotional states. This probability distribution vector is input into a threshold determination module. This threshold determination module generates an alarm signal when a specific emotional score exceeds a predetermined value. Based on this alarm signal, the unit automatically generates a coordination message. The unit multicasts the generated text and emotional data to various terminals. This process enables real-time cross-cultural communication without human interpreters. Batch processing on the server side reduces the computational load on the terminal side. This communication support unit is applied to multilingual international conference systems. This communications support department is used for daily stand-up meetings of multinational development teams. This communications support department is also used on global online discussion platforms.

[0043] The project schedule management is primarily implemented on the server side. The server accumulates user activity history (meeting participation history, speech content, task completion status, file sharing history, meeting minutes, chat logs, whiteboard entries, project milestone achievement status, each user's contribution, feedback history, questionnaire results, evaluation scores, etc.) and past project data as long-term memory within the virtual space. Generative AI analyzes this massive amount of data. For example, when signs of schedule delays are detected, it proposes reassessing task allocation or adding resources, increasing communication frequency, automatically reminding users of progress reports, automatically extracting delay factors, predicting risk factors, and automatically generating schedule charts or Gantt charts. Furthermore, when signs of cultural friction are detected (frequent negative emotions in interactions between specific members, a persistent atmosphere where members from specific countries or cultural spheres find it difficult to speak, or a lack of progress in reaching consensus in meetings, etc.), the AI ​​suggests arbitration or coordination proposals, such as "Cultural differences may affect the discussion. Let's organize the opinions of each country in order," or "In similar past projects, consensus-building processes like ○○'s have been effective," supporting the smooth operation of the team. In the early stages of the project, AI extracts successful patterns or failed cases from similar past projects and can propose optimal scheduling or risk management strategies, task decomposition schemes, role sharing schemes, communication rules schemes, meeting system schemes, and progress report format schemes. Furthermore, AI analyzes each user's skills, past contributions, professional fields, areas of expertise, past evaluation scores, and recommendation history from other members to automatically propose optimal task allocation or talent configuration, mentor-student combinations, leadership candidate recommendations, and optimal team composition schemes for each project. Specifically, the project schedule management department implements Long Short-Term Memory Networks or Time Series Transformers specifically designed for time series data analysis. The project schedule management department regularly extracts activity log data from the database. The project schedule management department uses, for example, a time series array of task completion indicators from the past 30 days as input data. The project schedule management department uses, for example, a normalized numerical vector representing the frequency of each member's speech as input data. The project schedule management department uses, for example, a moving average tensor of sentiment scores extracted from chat logs as input data. The project schedule management department combines these input data along the time axis to generate a multidimensional time series tensor. The project schedule management department inputs this multidimensional time series tensor into a time series Transformer. This time series Transformer calculates the attention weights between past patterns and the current state. The project schedule management department outputs, for example, a project delay probability score for the next week. The project schedule management department also outputs, for example, an importance vector of a specific communication pattern that could be a cause of cultural friction.The project schedule management department outputs, for example, a ranking matrix representing the task suitability of each member. This output is then fed to a rule-based inference engine. This inference engine triggers a resource reallocation algorithm when the delay probability score exceeds a threshold. The project schedule management department uses this algorithm to generate optimal task allocation change plans. This process allows for the early detection of minor schedule delays or team discord that are difficult for human managers to perceive. The project schedule management department quantitatively improves the project's success probability through a data-driven approach. This project schedule management department applies this tool to the management of large-scale software development projects. This project schedule management department applies it to the schedule management of multinational marketing campaigns. This project schedule management department applies it to projects building international supply chains.

[0044] On the terminal side, it is responsible for displaying the user interface, receiving user input, displaying real-time translation or sentiment analysis results, operating virtual avatars within the virtual space, uploading and downloading files, receiving notifications, operating the whiteboard or chat, joining and leaving meetings, answering questionnaires, providing feedback, and approving and rejecting AI proposals. For example, when a user enters a virtual meeting room from the terminal, the server sends the meeting room layout information, participant list, translated and culturally adapted message history, sentiment visualization information, progress chart, task overview, meeting agenda, file sharing links, and whiteboard content to the terminal, which displays this content in real time. When a user speaks, the terminal sends the voice data to the server, where generative AI performs translation, cultural adaptation, and sentiment recognition processing before distributing optimized messages or voice data to each recipient terminal. For example, content spoken in Chinese during a meeting is simultaneously displayed on each user terminal in English, Japanese, and Spanish, conveyed in a way that conforms to each culture. When a user writes diagrams or notes on the whiteboard, the content is automatically multilingualized, allowing members from different countries to understand in their native languages. Specifically, this terminal functions as an edge computing device. This terminal uses its built-in graphics processing unit to perform rendering processing. It uses, for example, two-dimensional coordinate vectors from the touchscreen as input data. It uses, for example, RGB image tensors at 30 frames per second acquired from the built-in camera as input data. It uses, for example, multi-channel speech waveform data acquired from a microphone array as input data. This terminal applies a lightweight neural network on the edge device to these input data. This lightweight neural network uses, for example, the MobileNet architecture. This terminal outputs, for example, a three-dimensional vector representing the user's gaze direction as output data. This terminal outputs, for example, a joint coordinate matrix representing the user's skeletal posture as output data. This terminal outputs, for example, a clean speech feature vector with background noise removed as output data. This terminal sends these output data as differential data to the server. This terminal unfolds polygon and texture data received from the server into local memory. This terminal executes a shader program to render virtual space onto the display. This terminal overlays translated text data at specified coordinates on the screen. This terminal distributes the server's computational load through this edge-side preprocessing and rendering process. This terminal reduces the amount of communication data with the server to save network bandwidth. This terminal thus suppresses the response latency to user operations to the millisecond level. This terminal is used in mobile devices such as smartphones or tablets. This terminal is also used in virtual reality headsets or augmented reality glasses. This terminal can be used with desktop personal computers or dedicated video conferencing terminals.

[0045] Furthermore, generative AI also learns from user behavior patterns, project progress, team dynamics, past successes and failures, user feedback or evaluations, external business news or industry trends, holiday or event information in various countries, current events, laws and regulations, exchange rates, or logistical conditions, contributing to the optimization of the entire system. For example, by learning the tendency for communication to become more active during specific time periods, AI can automatically propose schedules for important meetings or tasks, or automatically adjust project milestones considering holidays or large events in various countries, or predict the risks of exchange rate fluctuations or logistical delays and suggest supply chain adjustments. In addition, AI collects user feedback or evaluations to continuously improve algorithms for translation accuracy or cultural adaptation. For example, when users input feedback such as "This translation is unnatural," "This expression is impolite in my culture," or "This AI proposal does not fit the actual situation," AI learns the content and reflects it in the next and subsequent translations or proposals. Specifically, this system has a reinforcement learning agent and a fine-tuning module. This system executes a continuous learning pipeline in the background. This system uses, for example, paired data of translated text corrected by users as input data. This system uses, for example, a five-stage evaluation score vector of AI proposals as input data. It also uses, for example, embedding tensors of text data scraped from external news websites as input data. This system transforms these input data into a training dataset through batch processing. This system uses the training dataset to fine-tune the weight parameters of a large language model. This system uses a low-rank adaptation method to efficiently update only a subset of the model's parameters. This system outputs, for example, a difference matrix of the updated weight parameters as output data. This system outputs, for example, optimized parameters of the reward function of a scheduling algorithm as output data. This system outputs, for example, a coefficient vector of a new risk prediction model reflecting changes in the external environment as output data. This system dynamically loads this difference matrix into the inference server of the production environment. This process improves the performance of AI models without system downtime. This system improves the accuracy of personalization by converting implicit user feedback into explicit learning signals. This system utilizes computer technology to achieve system adaptability to changes in the external environment. This system is applied to market forecasting and team collaboration systems in global financial institutions. This system is applied to supply chain optimization platforms in multinational manufacturing industries. This system is applied to the knowledge management system of international consulting firms.

[0046] Thus, in this embodiment of the invention, by primarily implementing the functions of the virtual space generation department, communication support department, and project progress management department on the server side, and handling the user interface, real-time display processing, and user input reception on the terminal side, a highly cross-border team collaboration environment centered on generative AI is provided. For example, in all cross-border and cross-cultural teams such as global new business development projects, international R&D teams, startups primarily operating remotely, large corporations with multinational branches, international freelance groups, international conferences or academic societies, global education projects, international volunteer activities, cross-border supply chain management, international marketing activities, overseas branch launch projects, international legal and intellectual property teams, and global customer support centers, efficient and smooth collaboration can be achieved across physical, linguistic, and cultural barriers. Furthermore, through the automatic learning and optimization of AI, the system continuously evolves, flexibly and continuously responding to diverse user needs or changes in the global environment. Specifically, this system adopts a cloud-native microservice architecture. Each functional module is deployed as an independent container. The system uses time-series metrics such as the overall system's CPU utilization or memory consumption as input data. The system uses, for example, a network traffic matrix representing the communication latency between microservices as input data. This system uses, for example, a predicted vector of the number of concurrently connected users as input data. This input data is fed into a resource allocation optimization model. This resource allocation optimization model executes a deep reinforcement learning algorithm. The system outputs, for example, server resource allocation ratios for each container. The system outputs, for example, control commands instructing the horizontal scaling of the inference server. The system outputs, for example, change parameters for database replication settings. The system drives a container orchestration tool based on these outputs. Through this processing, the system maintains stability even in the face of traffic surges. The system minimizes operating costs by releasing unnecessary server resources. The system improves the availability of the entire computer system through dynamic and autonomous management of hardware resources. This system is applied to virtual platforms for large-scale international events requiring tens of thousands of simultaneous connections. This system is applied to collaborative environments for open-source projects involving thousands of developers distributed around the world. This system is applied to remote diagnostics and conferencing systems conducted by multinational medical expert teams.

[0047] (System Composition) The system described in this implementation includes a virtual space generation department, a communication support department, a project progress management department, a user interface department, a database management department, a security management department, and a generative AI collaboration department. The virtual space generation department receives various attribute information from user terminals, such as name, affiliation, language, cultural background, time zone, job title, professional field, past project participation history, desired work style, virtual avatar settings, and preferred interface design, as user information. Based on this information, it generates a virtual office space optimized for each user. For example, in the case of a company with branches in multiple regions such as Asia, Europe, North America, South America, Africa, and Oceania, it integrates the physical location information and network information of each branch to construct a metaverse space that virtually connects all branches. Within the virtual space, various areas are set up, such as meeting rooms, individual cubicles, lounges, project rooms, a cafeteria, an activity hall, a library, a rest area, a reception area, a whiteboard area, and a presentation room. Users can freely move between these areas from their terminals. For example, it can also automatically generate dedicated virtual meeting rooms or group discussion rooms for each project and dynamically optimize the layout and interface based on the participants' attributes. Furthermore, it can automatically switch the lighting and background music, avatar clothing, and display language within the virtual space based on the user's time zone and work period. Specifically, this virtual space generation unit runs on a dedicated server instance equipped with hardware accelerators. This virtual space generation unit uses, for example, structured data in JSON format containing user attribute information as input data. This virtual space generation unit uses, for example, geospatial coordinate vectors containing latitude and longitude information of each network point as input data. This virtual space generation unit uses, for example, a numerical array in milliseconds representing network latency as input data. This virtual space generation unit inputs these input data into a multilayer perceptron to convert them into latent spatial vectors. This virtual space generation unit inputs these latent spatial vectors into a conditional generation model. This conditional generation model employs a variational autoencoder architecture. This virtual space generation unit outputs, for example, an adjacency matrix representing the virtual space topology as output data. This virtual space generation unit outputs, for example, coordinate tensors representing the dimensions of each room's 3D bounding box as output data. This virtual space generation unit outputs, for example, a lighting parameter vector representing the color temperature or intensity of ambient light as output data. This virtual space generation unit dynamically constructs a scene graph based on this output data. This virtual space generation unit distributes the scene image to each terminal, instructing them to render it locally. Through this process, it provides personalized spaces while avoiding heavy server-side graphics processing. It achieves a rich visual experience while minimizing network bandwidth consumption. This virtual space generation unit is used in collaborative production studios for multinational creative teams. It is used in virtual property navigation systems for global real estate companies. It is used in virtual viewing arenas for international esports tournaments.

[0048] The Communications Support Department has the capability to translate messages in real time, including text chat, voice calls, video conferencing, whiteboard sharing, file sharing, and emotional expressions via stickers or emoticons. For example, it can simultaneously translate Japanese messages into multiple languages ​​such as English, Chinese, Spanish, French, German, Arabic, and Hindi, and distribute them to various user terminals. The translation process uses a large-scale language model based on generative AI and supports specialized terminology, industry-specific expressions, abbreviations, slang, dialects, business terms, company internal language, and project-specific abbreviations. For instance, in IT projects, terms such as "API," "deployment," and "Agile" are translated into appropriate specialized terms in each language. Furthermore, the system automatically adapts to different cultures by considering each user's cultural background, past communication history, business etiquette, religious considerations, gender expressions, distinctions between honorific and informal language, interpretations of humor or sarcasm, and the meanings of nonverbal expressions (gestures or facial expressions). For example, it converts honorific expressions from Japanese culture to informal expressions from American culture, adjusts euphemisms from European culture to direct expressions from Asian culture, or converts expressions to reflect religious considerations within the Islamic cultural sphere. For instance, if an Indian member says "We'll discuss it," it can convey the clear intention to an American member as "I'll get back to you soon." Additionally, using generative AI-based emotion recognition algorithms, the system analyzes positive, negative, stressful, tense, reassured, angry, confused, expectant, joyful, disappointed, and fatigued emotional states from tone of voice, facial expressions, gestures, intonation, speaking speed, typing speed, and the frequency of emoji or sticker usage. These emotions are then visualized on the terminal screen using icons, color differentiation, charts, heatmaps, and changes in the virtual avatar's facial expressions. For example, when some members experience increased stress during a meeting, the AI ​​automatically inserts a coordinating message, such as "The discussion seems intense. We suggest a 5-minute break," or "Opinions seem to differ. Let's clarify our positions," to ease the discussion or propose a break. Furthermore, the AI ​​analyzes users' speaking frequency, speaking time, the diversity of their content, and the number of responses to others. For members who speak less, it can automatically generate encouraging messages such as "We'd also like to hear Mr. / Ms. ○'s opinion." Specifically, this communications support department uses a stream processing engine to perform pipelined data processing. This communications support department uses, for example, spectrogram tensors of real-time sampled speech as input data. This communications support department uses, for example, character code arrays from text chat as input data. This communications support department uses, for example, time-series vectors in milliseconds representing keystroke intervals as input data. This communications support department feeds this input data into a multimodal integration model. This multimodal integration model maps speech, text, and action data into a common latent space.This communication support unit outputs, for example, a Unicode string of the translated text as output data. This communication support unit outputs, for example, an output vector representing the probability of each sentiment category using a Softmax function as output data. This communication support unit outputs, for example, a binary flag that serves as a trigger for coordinating message generation as output data. This communication support unit distributes this output to each terminal via a publish-subscribe messaging system. This communication support unit minimizes translation latency for a large number of simultaneously connected users through this asynchronous processing architecture. This communication support unit does not mimic human cognitive processes, but achieves cultural adaptation through mathematical operations in a high-dimensional vector space. This communication support unit is applied to multilingual online customer support systems. This communication support unit is applied to international telemedicine consultation platforms. This communication support unit is applied to online language learning systems involving multinational students.

[0049] The Project Schedule Management Department accumulates user activity history (meeting participation history, speech content, task completion status, file sharing history, meeting minutes, chat logs, whiteboard entries, project milestone achievement status, each user's contribution, feedback history, questionnaire results, evaluation scores, etc.) and past project data in the virtual space as long-term memory, and uses generative AI to analyze this massive amount of data. For example, when signs of schedule delays are detected, proposals are made to reassess task allocation or add resources, increase communication frequency, automatically remind users of progress reports, automatically extract delay factors, predict risk factors, and automatically generate schedule charts or Gantt charts. In addition, when signs of cultural friction are detected (frequent negative emotions in interactions between specific members, a persistent atmosphere where members from specific countries or cultural circles find it difficult to speak, or no progress in reaching consensus in meetings, etc.), AI suggests arbitration or coordination proposals, such as "Cultural differences may affect the discussion. Let's organize the opinions of each country in order." or "In similar projects in the past, consensus-building processes like ○○'s have been effective." These proposals support the smooth operation of the team. In the early stages of the project, AI extracts successful patterns and failure cases from similar past projects, and can also propose optimal scheduling or risk management strategies, task decomposition schemes, role sharing schemes, communication rules schemes, meeting system schemes, progress report format schemes, etc. Furthermore, AI analyzes each user's skills and past contributions, professional fields, areas of expertise, past evaluation scores, and recommendation history from other members, automatically proposing optimal task allocation and talent configuration, mentor-student combinations, leadership candidate recommendations, and optimal team composition schemes for each project. Specifically, the project's progress management department adopts a Lambda architecture integrating batch processing and stream processing. The project's progress management department uses, for example, directed acyclic graph data representing task dependencies from past projects extracted from a database as input data. The project's progress management department uses, for example, multidimensional binary vectors representing each member's skill set as input data. The project's progress management department uses, for example, floating-point arrays representing real-time task utilization as input data. The project's progress management department feeds these input data into a graph convolutional network. This graph convolutional network learns the relationships between nodes and extracts features. The project schedule management department outputs, for example, a probability distribution tensor representing the expected completion date of each task. It also outputs, for example, a vector representing the network centrality scores of communication bottlenecks within the team. Furthermore, it outputs, for example, an allocation matrix representing the optimal member combination. These outputs are then passed to a dashboard generation module to automatically generate visual reports. Through this processing, the project schedule management department supports data-driven decision-making that does not rely on the experience and intuition of project managers.The Project Schedule Management Department uses comprehensive computer-based analysis of large-scale data to identify risk factors that are difficult for humans to detect. This department is applied to the engineering management system of large-scale infrastructure construction projects. It is also applied to the production management of multinational film production teams. Furthermore, it is used in the schedule tracking system of an international collaborative research alliance.

[0050] The user interface provides a multilingual and multicultural interface on the terminal, enabling users to operate virtual avatars within the virtual space, moving them to meeting rooms, lounges, whiteboard areas, etc., or intuitively engaging in chat, voice calls, video conferencing, file sharing, sending stickers, answering questionnaires, providing feedback, and approving or rejecting AI proposals. For example, when a user enters a virtual meeting room from the terminal, the server sends the meeting room layout information, participant list, translated and culturally adapted message history, emotional state visualization, progress chart, task overview, meeting agenda, file sharing links, and whiteboard content to the terminal, which displays this content in real time. When a user speaks, the terminal sends the voice data to the server, where generative AI performs translation, cultural adaptation, and emotion recognition processing before distributing optimized messages or voice data to each recipient terminal. For example, content spoken in Chinese during a meeting is simultaneously displayed on each user terminal in English, Japanese, and Spanish, conveyed in expressions appropriate to each culture. When users write diagrams or notes on the whiteboard, the content is automatically multilingualized, allowing members from different countries to understand in their native languages. Specifically, this user interface is implemented as a dedicated client application running on the terminal's operating system. This user interface uses time-series arrays of input events, such as those from a mouse or keyboard, as input data. This user interface uses two-dimensional tensors, such as those from stylus pressure and coordinates, as input data. This user interface uses three-dimensional acceleration vectors, such as those from a gyroscope sensor, as input data. This user interface processes this input data in a local event loop. This user interface outputs, for example, a two-dimensional array of pixel colors drawn on the screen as output data. This user interface outputs, for example, a movement vector of a virtual avatar as output data. This user interface outputs, for example, serialized command data packets sent to a server as output data. This user interface directly calls the terminal's graphics API for hardware-accelerated rendering. Through this processing, this user interface achieves smooth screen updates at high frame rates. This user interface prevents screen freezing caused by network latency by communicating asynchronously with the server. This user interface thereby enhances user immersion and provides intuitive operability. This user interface is used as a collaborative design tool for multinational design teams. This user interface is used as an interactive whiteboard on an international online education platform. This user interface is used as the control console for the global remote surgical support system.

[0051] The Database Management Department centrally manages and stores user information, virtual space composition information, communication history, project progress data, file sharing data, AI-generated analysis results, feedback history, questionnaire results, evaluation scores, past project history, and various log data, enabling high-speed retrieval, extraction, and aggregation as needed. For example, it can retrieve past meeting minutes or chat history to extract case studies for similar situations, aggregate each user's contribution or speaking tendencies for use in talent evaluation or task allocation, or efficiently provide large amounts of data for AI-driven progress analysis or risk prediction. Specifically, this Database Management Department is constructed by combining a distributed relational database and a vector database for unstructured data. It uses string data, such as text documents, as input data. It uses binary data, such as image files, as input data. It uses structured JSON data, such as system logs, as input data. The department inputs text or image data into an embedding model, converting it into a high-dimensional dense vector. This dense vector is then indexed and stored in the vector database. This database management unit outputs, for example, a descending list of cosine similarity scores for a retrieval query. It also outputs, for example, a set of records that meet specific criteria. Furthermore, it outputs, for example, a tensor of statistical values ​​calculated by a summarization function. Using an approximate nearest neighbor search algorithm, it retrieves similar data from millions of records in milliseconds. Through this processing, it provides high-speed contextual data for retrieval enhancements generated for large language models. It ensures high availability and fault tolerance through data sharding and replication. This database management unit is used in company-wide knowledge management systems for multinational corporations. It is used in international patent investigation and analysis platforms. It is used in comprehensive retrieval systems for global clients' historical responses.

[0052] The Security Management Department performs tasks such as user authentication, access control management, communication encryption, data backup, information leakage countermeasures, unauthorized access detection, compliance with personal information protection regulations in various countries, and audit log management to maintain a secure environment that allows multinational teams to use the system with peace of mind. For example, it sets detailed access permissions for each project, restricting access to specific confidential information or files; it detects and blocks unauthorized access or malware attacks from external sources in real time; and it automatically optimizes data storage destinations or data transmission paths according to the regulations of various countries. Specifically, this Security Management Department functions as an authentication and authorization proxy server based on a zero-trust architecture. This Security Management Department uses, for example, multidimensional vectors containing packet header information of network traffic as input data. This Security Management Department uses, for example, arrays of timestamps and IP addresses of user login attempts as input data. This Security Management Department uses, for example, tensors of statistical features extracted from the payload as input data. This Security Management Department feeds this input data into an autoencoder for anomaly detection. This autoencoder learns normal communication patterns and reconstructs the input data after compressing it to a low-dimensional space. This Security Management Department outputs, for example, a reconstruction error score between the input data and the reconstructed data. This security management department outputs, for example, binary judgment values ​​for allowing or denying access requests as output data. This security management department outputs, for example, routing labels specifying the region where the data is stored. This security management department automatically terminates the session if the reconstruction error score exceeds a threshold. Through this processing, this security management department detects and defends against unknown cyberattacks or internal violations in real time. This security management department implements dynamic data routing that complies with national data localization regulations through computer processing. This security management department is used in secure information sharing platforms for multinational financial institutions. This security management department is used in collaborative development networks for the international defense industry. This security management department is used in clinical trial data management systems for global pharmaceutical companies.

[0053] The generative AI collaboration unit receives requests from various units including the virtual space generation unit, communication support unit, and project progress management unit, generates appropriate prompts and sends them to the generative AI, and feeds the response results from the AI back to each unit. For example, in response to a request from the virtual space generation unit, a prompt stating "Generate a virtual office space for a multinational team with 3 sites in Asia, Europe and North America, whose layout reflects the cultural characteristics of each site" is generated and sent to the AI. In addition, in response to a request from the communication support unit, a prompt stating "Translate the Japanese utterance 'ご検討のほどよろしくお願いいたします' into an informal English expression aligned with American culture" is generated and sent to the AI. In response to a request from the project progress management unit, prompts such as "Analyze the project progress data and communication history of the past 6 months, and propose the main factors of schedule delays and improvement strategies" and "Generate a coordination proposal for alleviating cultural frictions between specific members" are generated and sent to the AI. Furthermore, based on feedback or evaluations from users, prompts such as "Modify this translation result to a more natural expression" and "Summarize the minutes of this meeting, and extract action items that take the cultural backgrounds of various countries into consideration" are also generated. Specifically, this generative AI collaboration unit functions as an API gateway and a prompt engineering module. This generative AI collaboration unit uses, for example, key-value pairs of context variables transmitted from each unit as input data. This generative AI collaboration unit uses, for example, a state vector representing the current state of the user as input data. This generative AI collaboration unit uses, for example, a token array of past conversation history as input data. This generative AI collaboration unit applies these input data to a template engine to dynamically synthesize a structured prompt string. This generative AI collaboration unit inputs the prompt string into a tokenizer to convert it into an integer index array. This generative AI collaboration unit outputs, for example, a token embedding tensor suitable for the input layer of a large language model as output data. This generative AI collaboration unit outputs, for example, the payload of an HTTP request to an inference server as output data. This generative AI collaboration unit outputs, for example, extracted text data after parsing the response JSON from the AI as output data. This generative AI collaboration unit performs connection pooling in communication with the inference server to reduce communication overhead. Through this processing, this generative AI collaboration unit enables each module in the system to utilize advanced inference functions without being aware of the complex internal structure of the AI model. This generative AI collaboration unit stabilizes the output accuracy of the AI and suppresses hallucinations through prompt optimization. This generative AI collaboration unit is applied to an automatic contract review system for multinational law firms. This generative AI collaboration unit is applied to an automatic multilingual article generation system for international news media. This generative AI collaboration unit is applied to an employee engagement analysis system for global human resources departments.

[0054] Examples of specific prompt texts may include "Please translate the Japanese utterance 'ご検討のほどよろしくお願いいたします' into an informal English expression adapted to American culture", "Please convert honorific expressions used in business meetings in Chinese culture to direct expressions adapted to German culture", "Please analyze the project progress data and meeting logs of the past 3 months, identify factors causing schedule delays, and propose 3 improvement strategies", "Please summarize the minutes of this meeting and extract action items that take into account the cultural backgrounds of various countries", "Please detect signs of cultural friction from chat histories between specific members and propose mitigation strategies", "Please generate a virtual office space for a multinational team with three sites in Asia, Europe and North America, the layout of which reflects the cultural characteristics of each site", and the like. Specifically, the present system internally converts these natural language-based prompt texts into machine-readable vector representations for processing. The present system uses, for example, the original string of the prompt text as input data. The present system uses, for example, text data of constraint conditions set as system prompts as input data. The present system uses, for example, paired arrays of example data for few-shot learning as input data. The present system optimizes these input data using a prompt fine-tuning module. The present system outputs, as output data, for example, a soft prompt tensor for guiding the attention mechanism of a large language model to a specific context. The present system outputs, as output data, for example, a penalty parameter for controlling the length of generated text. The present system outputs, as output data, for example, a numerical value of a temperature parameter for adjusting the diversity of outputs. The present system appends these parameters to an inference request and transmits the request to the large language model. Through this processing, the present system realizes mathematical control of contexts in a high-dimensional space, instead of simple string transmission. The present system thereby guarantees stable inference results that accurately reflect human intentions as a computer technology. The present system is applied to automatic response script generation for multinational customer support. The present system is applied to qualitative data analysis for international market research. The present system is applied to multilingual requirement specification generation in global software development.

[0055] Thus, the system described in this embodiment achieves an advanced virtual office environment by organically collaborating the virtual space generation department, communication support department, project schedule management department, user interface department, database management department, security management department, and generative AI collaboration department. This enables multinational teams to collaborate smoothly and efficiently across physical, linguistic, and cultural barriers. Specifically, this system constructs a comprehensive data processing pipeline that integrates modules through a high-speed internal bus and message queues. The system uses, for example, multimodal streaming data continuously flowing in from a group of terminals as input data. The system uses, for example, petabytes of historical data accumulated in a database as input data. The system uses, for example, real-time environment variable vectors obtained from external APIs as input data. The system processes these input data in parallel and in a distributed manner across modules. The system outputs, for example, a synchronized global state matrix of the virtual space as output data. The system outputs, for example, multilingual event streams distributed to all user terminals as output data. The system outputs, for example, a summary tensor representing the overall system's operational status for a dashboard as output data. Through this organic collaboration, the system solves the complex cross-cultural communication challenges that cannot be addressed by a single module. This system ensures high-speed and scalable processing by replacing the translation or coordination of human mental activities with the physical computation of computers. It thus provides a next-generation collaborative foundation that surpasses existing video conferencing systems. This system is applied to integrated control systems for multinational space development projects. It serves as a real-time information-sharing platform for international disaster relief teams. It is also used in multi-institutional collaborative simulation environments for global climate change research.

[0056] (Implementation steps) Step 1: User Information Registration and Virtual Space Initialization Users input detailed attribute information from their terminals, including name, affiliation, language, cultural background, time zone, job title, field of expertise, past project experience, preferred work style, virtual avatar settings, and preferred interface design, and send this information to the server. The server receives this information and initializes a virtual office space optimized for each user. For example, if a server has locations in multiple regions across Asia, Europe, North America, South America, Africa, and Oceania, it integrates the physical location and network information of each location to generate a metaverse space that virtually connects them. Within this virtual space, various areas are set up, such as meeting rooms, individual cubicles, lounges, project rooms, a cafeteria, an activity hall, a library, a rest area, a reception area, a whiteboard area, and a presentation room. Specifically, the server encodes user attributes into multi-dimensional profile vectors during registration processing. The server uses, for example, text strings entered by the user in a form as input data. The server uses, for example, the ID of a categorical variable selected from a dropdown menu as input data. The server uses, for example, the pixel tensor of an uploaded virtual avatar image as input data. The server normalizes and vectorizes this input data in a preprocessing module. This server outputs, for example, hash values ​​of user records stored in a database. It also outputs, for example, JSON data defining the initial layout of a virtual space. Furthermore, it outputs, for example, 3D mesh data representing the initial state of a virtual avatar. This server minimizes user login latency through its initialization algorithm. It achieves personalization while minimizing computational resource consumption by combining pre-calculated templates. This server is used in virtual office wizards within multinational corporations' onboarding systems. It is used in profile registration systems for exhibitors and visitors at international online exhibitions. And it is used for generating virtual dating spaces in global matching applications.

[0057] Step 2: Dynamic Optimization of Virtual Space The server dynamically optimizes the layout, display language, lighting, background music, and avatar clothing within the virtual space based on users' time zones, work hours, projects, positions, professional fields, and past activity history. For example, during periods with more users in the Asian region, it automatically sets decorations or background music that consider Asian culture; when there are more users in the European region, it performs spatial performances that match European holidays or events. Specifically, this server uses a real-time inference engine to dynamically update environmental parameters. The server uses, for example, a set of profile vectors for all currently logged-in users as input data. The server uses, for example, numerical data of the current time based on Coordinated Universal Time (UTC) as input data. The server uses, for example, the coordinate matrix of each user's current virtual space as input data. The server feeds this input data into a clustering algorithm to analyze the user distribution. The server outputs, for example, a color vector specifying the RGB values ​​of the ambient light in the virtual space as output data. The server outputs, for example, a list of URIs for the background music audio files as output data. The server outputs, for example, the vertex coordinate tensors of dynamically generated 3D objects as output data. Based on these outputs, the server updates the state machine of the virtual space and pushes and distributes difference data to each terminal. This server avoids re-rendering the entire space through this process, saving network bandwidth. This server provides an adaptive interface that responds immediately to changes in user conditions using computer technology. This server is used in dynamic map generation systems for cross-border online games. This server is used in time-based store layout change systems for international virtual marketplaces.

[0058] Step 3: Multilingual and Multicultural Communication Support Users use various communication means in the virtual space, including text chat, voice call, video conference, whiteboard sharing, file sharing, and emotional expression via stickers or emojis. The server uses generative AI to perform real-time multilingual translation on messages between users, and automatically adjusts cultural background, past communication history, business etiquette of various countries, religious considerations, gender expression, the differential use of honorific and informal expressions, interpretation of humor or irony, the meaning of non-verbal expressions, etc. For example, prompt sentences such as "Please translate the Japanese utterance 'ご検討のほどよろしくお願いいたします (looking forward to your discussion)' into an informal English expression adapted to American culture" and "Please convert the honorific expression used in business meetings in Chinese culture into a direct expression adapted to German culture" are sent to the generative AI. Specifically, this server executes a streaming translation pipeline at high speed in the memory. This server uses, for example, a text token stream sequentially output from a speech recognition engine as input data. This server uses, for example, a multidimensional vector representing the cultural attributes of the sender and the receiver as input data. This server uses, for example, a context embedding tensor of the recent dialogue history as input data. This server inputs these input data into the encoder of a large language model to calculate a latent representation. This server outputs, for example, a text token sequence in the target language as output data. This server outputs, for example, an array of probability scores representing translation confidence as output data. This server outputs, for example, an index array representing the highlighted positions of words modified by cultural adaptation processing as output data. When the translation result reaches a certain length, this server sends it to each terminal in units of blocks. Through this streaming processing, this server displays the translation result without waiting for the end of the sentence, reducing the communication delay to the limit. Through vector operations in a high-dimensional space, this server automates complex cultural adaptation that cannot be achieved based on rules. This server is applied to an online support system for cross-border diplomatic negotiations. This server is applied to a real-time caption generation system for international news agencies. This server is applied to a multilingual simultaneous interpretation platform in global online education.

[0059] Step 4: Emotion recognition and communication visualization The server uses a generative AI-based emotion recognition algorithm to analyze emotional states such as positive, negative, stress, tension, peace of mind, anger, confusion, anticipation, joy, disappointment, and fatigue from factors such as tone of voice, facial expressions, gestures, intonation, speaking speed, chat input speed, and the frequency of emoji or sticker usage in messages or speeches. These emotions are visualized on the terminal screen using icons, color differentiation, charts, heatmaps, and changes in the expressions of virtual avatars. As an example of applying generative AI, one could ask, "Analyze the emotional states of each participant in the chat logs and voice data of this meeting in a time-series manner, and determine the timing of increases in stress or tension." Specifically, this server uses a multimodal neural network to comprehensively analyze data from multiple sensors. The server uses, for example, a time-series feature matrix of pitch or power extracted from speech waveforms as input data. The server uses, for example, motion unit vectors representing facial muscle movements extracted from camera images as input data. The server uses, for example, an array of emotion scores representing the emotional polarity of the text as input data. This server feeds the input data into a fusion layer with an attention mechanism, dynamically weighting the importance of each modality. The server outputs, for example, a Softmax vector representing the probability distribution of ten basic emotions. The server outputs, for example, a continuous score representing stress levels from zero to one hundred. The server outputs, for example, a tensor representing the blending deformation parameters used to control the facial expressions of virtual avatars. The server stores these outputs in a time-series database and sends them in real-time to the rendering engines of each terminal. Through this processing, the server extracts and visualizes subtle nonverbal changes in humans as quantitative data. The server technically prevents communication errors in online communication by objectively quantifying emotions. This server is used in online consultation systems for cross-border mental health care. This server is used in assessment support tools for international personnel interviews. This server is used in operator stress management systems for global call centers.

[0060] Step 5: AI-assisted coordination and cultural friction mitigation The server uses generative AI to analyze users' speaking frequency, speaking time, diversity of speaking content, and number of responses to others. For members who speak less, it automatically generates encouraging messages such as "We'd also like to hear Mr. / Ms. ○'s opinion." Additionally, when signs of cultural friction are detected, it suggests coordination solutions such as "Differences in cultural backgrounds may affect the discussion. Let's organize the opinions of each country in order." As an example of providing feedback to generative AI, one could be, "Please detect signs of cultural friction from the meeting logs of the past month and propose three mitigation strategies." Specifically, this server executes a natural language processing model. This model uses a Transformer architecture with a self-attention mechanism. This server converts each user's speech data into text data. This server converts the text data into word embedding vectors. This server extracts speaking time or speaking frequency as numerical vectors. This server combines these vectors to generate a multi-dimensional input tensor. For example, the input tensor is a 512-dimensional floating-point array. This model processes the input tensor. This model outputs a score representing the probability of cultural friction occurring. For example, the output is a floating-point number in the range of 0.0 to 1.0. This server determines whether the output score exceeds a preset threshold. If it does, the server generates a coordination message. This message is then sent to each terminal. Thus, this system achieves objective friction detection independent of human subjective judgment. This system enables real-time intervention while suppressing unnecessary consumption of communication bandwidth. This technology is applied to online conferencing systems for multinational teams.

[0061] Step 6: Automated Management and Optimization Proposals for Project Schedule The server accumulates user activity history, meeting participation history, speech content, task completion status, file sharing history, meeting minutes, chat logs, whiteboard entries, project milestone achievement status, each user's contribution, feedback history, questionnaire results, and evaluation scores as long-term memory within the virtual space. It uses generative AI to automatically extract signs or risk factors for progress delays, reassess task allocation, propose additional resources, automatically remind users of progress reports, automatically generate progress charts or Gantt charts, and recommend leadership candidates. As examples of suggestions for generative AI, one could ask, "Please analyze the project progress data and communication history over the past 6 months, propose the main factors causing progress delays and improvement strategies," or "Please generate the optimal task allocation plan based on the project's member composition and past contributions." Specifically, this server executes a recurrent neural network for time-series data analysis. The server converts activity history or task completion status into a time-series numerical array. The server converts natural language meeting minutes into sentence embedding vectors. The server integrates this data to construct an input tensor. For example, the input tensor is a 3D tensor representing data from the past 30 days. This network processes the input tensor. The network outputs a score vector representing the latency probability of each task. For example, the output is a latency risk score from 0 to 100 for each task. The server executes a resource reallocation algorithm based on the latency risk scores. The server generates an optimal task allocation scheme. The server stores the generated allocation scheme in a database. The server sends notification signals to the corresponding user terminals. Thus, this system achieves efficient allocation of computer resources in project management. This system reduces data processing latency. This technology is applied to management tools for large-scale software development projects.

[0062] Step 7: Information provision and operation via user interface On the terminal side, a multilingual and multicultural interface is provided, enabling users to operate virtual avatars within the virtual space, moving them to meeting rooms, lounges, whiteboard areas, etc., or intuitively engaging in chat, voice calls, video conferencing, file sharing, sending stickers, answering questionnaires, providing feedback, and approving or rejecting AI proposals. For example, when entering a meeting room, the server sends the meeting room layout information, participant list, translated and culturally adapted message history, emotional state visualization, progress chart, task overview, meeting agenda, file sharing links, and whiteboard content to the terminal, which displays this content in real time. Specifically, this terminal uses a graphics processing unit to render a 3D virtual space. This terminal receives layout information and a participant list from the server. This terminal expands the received data into local memory. This terminal interprets the emotional state visualization as color information vectors. This terminal performs texture mapping based on color information vectors on the polygonal mesh of the virtual avatar. This terminal inputs the translated message history into the text rendering engine. This terminal renders text at specified coordinates on the screen. This terminal detects user input as interrupt signals. This terminal converts input operations into command data packets. This terminal sends command packets to the server via asynchronous communication. This distributes the rendering load on the terminal side, minimizing the impact of network latency. This technology is applied to multilingual virtual office applications.

[0063] Step 8: Continuous application of security, data management, and AI learning The server performs user authentication, access control management, communication encryption, data backup, information leakage countermeasures, unauthorized access detection, compliance with personal information protection regulations in various countries, and audit log management, maintaining a secure environment that allows global multinational teams to use the system with peace of mind. Furthermore, it collects user feedback and evaluations to continuously learn and improve the translation accuracy, cultural adaptation processing, sentiment recognition algorithms, and progress analysis algorithms of its generative AI. Examples of suggestions for generative AI include "Please revise the translation to a more natural expression" and "Please summarize the meeting minutes and extract action items that consider the cultural backgrounds of different countries." Specifically, the server executes an encryption module. It applies advanced encryption algorithms to communication data. It records audit logs in a database using a hash chain structure. It executes a machine learning module. It collects user feedback as teacher data. It inputs the feedback text into a natural language processing model. It updates the model's weight parameters using backpropagation. It calculates the cross-entropy error as the loss function. It minimizes the loss function using gradient descent. It stores the updated weight parameters in memory. Thus, the system continuously improves the model's inference accuracy. This system optimizes the computational cost required for relearning. This technology is applied to highly confidential international collaborative research platforms.

[0064] (specific use case) For example, imagine a global project team with offices in three regions: Asia, Europe, and North America, advancing a new product development project. Members at each office have different languages, cultural backgrounds, time zones, and areas of expertise. The virtual space generation department receives information such as each member's name, affiliation, language, cultural background, time zone, position, area of ​​expertise, and past project experience. Based on this information, it generates a metaverse space that virtually connects the offices in the three regions. Within this virtual space, dedicated meeting rooms, break rooms, lounges, whiteboard areas, file-sharing areas, and progress management areas are automatically generated for each project. Members can freely move between these areas using their virtual avatars on their terminals. Specifically, the server executes a space generation algorithm. The server acquires each member's attribute information as structured data. The server converts the attribute information into feature vectors. The server inputs the feature vectors into a clustering algorithm. The server calculates the similarity between members. The server determines the initial configuration coordinates within the virtual space based on the similarity. The server measures the network latency in the three regions. The server dynamically allocates edge servers based on the latency measurement results. The server generates polygonal and texture data for the virtual space. The server streams the generated data to each terminal. Therefore, this system hides communication latency caused by physical distance. This system stabilizes the rendering frame rate in virtual space. This technology is applied to virtual headquarters systems for global enterprises.

[0065] The communication support unit uses generative AI to perform real-time multilingual translation when members use various communication means such as text chat, voice call, video conference, whiteboard sharing, and file sharing in a virtual conference room. For example, the Japanese expression "ご検討のほどよろしくお願いいたします" (expecting your discussion) is translated into an informal English expression adapted to American culture "Looking forward to your feedback", and conveyed to European members with the subtle nuance of "looking forward to your opinions". In addition, the generative AI refers to the cultural background of each member, past communication history, business etiquette of each country, religious considerations, differentiated use of honorific and informal expressions, meanings of non-verbal expressions, etc. For example, when a member in the Indian cultural sphere says "I will discuss it", the intention is clearly conveyed to North American members as "I'll get back to you soon.". Furthermore, using a generative AI-based emotion recognition algorithm, the emotional state of each member is analyzed from the intonation of messages or speeches, facial expressions, gestures, speech prosody, chat input speed, use frequency of emojis or stickers, etc., and visualized on the terminal screen through icons, color differentiation, expression changes of avatars, etc. For example, when the stress of some members increases during a meeting, the AI automatically inserts a coordination message such as "The discussion seems very intense. A 5-minute break is proposed.". Specifically, this server executes a large language model and an emotion recognition model in parallel. This server converts the input speech stream into an acoustic feature quantity tensor. For example, the acoustic feature quantity tensor is a two-dimensional array including Mel-frequency cepstral coefficients. This server converts text data into a one-dimensional array of token IDs. This server inputs these tensors into each model. The large language model outputs a token probability distribution corresponding to the target translation language. The emotion recognition model outputs a probability score for each emotion category such as joy or anger. For example, the output is a floating-point vector with joy being 0.8 and anger being 0.1. This server decodes the optimal word sequence from the probability distribution. This server generates control parameters for the avatar based on the emotion scores. This server synchronizes the translated text and the control parameters and transmits them to the terminal. Accordingly, this system prevents the loss of information transmission of semantics and emotions between multiple languages. This system maximizes the parallel execution efficiency of the processor in real-time processing. This technology is applied to international customer support systems.

[0066] The Project Progress Management Department accumulates data such as meeting participation history, speaking content, task completion status, file sharing history, meeting minutes, chat logs, whiteboard entries, project milestone achievement status, each user's contribution, and feedback history in the virtual space as long-term memory. Generative AI is then used to automatically extract signs or risk factors for progress delays. For example, in the event of a delay, the generative AI is prompted to "analyze the project progress data and communication history of the past 6 months, propose the main factors causing the delay, and suggest improvement strategies." The AI ​​then proposes specific improvement strategies such as "reassessing task sharing," "adding resources," and "increasing communication frequency." Furthermore, when signs of cultural friction are detected, the generative AI is prompted to "detect signs of cultural friction from the meeting logs of the past month and propose three mitigation strategies," suggesting coordination solutions such as "organizing opinions from different countries in order," "implementing a vote to reach consensus," and "holding workshops with shared cultural backgrounds." Specifically, this server executes a graph neural network. The server converts the interactions between users into a graph data structure composed of nodes and edges. The server assigns feature vectors such as task completion rate or speaking frequency to each node. This server inputs graph data into the network. Information propagates between computing nodes in this network. The network output represents a regression value indicating the delay risk for the entire project. For example, the output is a continuous value ranging from 0 to 1. This server activates the alarm generation module when the regression value exceeds a specified threshold. This server uses the alarm generation module to generate specific improvement strategy text. This server writes the generated text into the project management database. This addresses bottlenecks in specific task groups with complex dependencies in the early stages of the system. This system reduces the computational resources required for data retrieval and aggregation. This technology is applied to hardware design projects undertaken by multinational teams.

[0067] The user interface supports multilingual and multicultural interfaces, enabling users to operate virtual avatars within the virtual space, moving them to meeting rooms, lounges, whiteboard areas, etc., and intuitively engaging in chat, voice calls, video conferencing, file sharing, sticker sending, questionnaire responses, feedback input, and the approval and rejection of AI proposals. For example, when entering a meeting room, the server sends the meeting room layout information, participant list, translated and culturally adapted message history, emotional state visualization, progress chart, task overview, meeting agenda, file sharing links, and whiteboard content to the terminal, which displays this content in real time. Specifically, the terminal executes an event-driven user interface control program. The terminal monitors interrupt requests from input devices. The terminal converts the virtual avatar's movement into a three-dimensional coordinate displacement vector. The terminal sends the displacement vector to the server. The terminal stores the layout information and participant list received from the server in a local cache. The terminal reads data from the cache and transmits it to the rendering buffer. The terminal applies the emotional state visualization as parameters for the particle system. The terminal renders the progress chart and task overview as two-dimensional vector graphics. Therefore, this system optimizes memory usage on the terminal side. This system also reduces processing latency associated with screen redrawing. This technology is applied to virtual collaboration tools for remote work.

[0068] The database management department centrally manages and stores user information, virtual space composition information, communication history, project progress data, file sharing data, AI-generated analysis results, feedback history, questionnaire results, evaluation scores, past project history, and various log data, enabling high-speed retrieval, extraction, and aggregation as needed. For example, it can retrieve past meeting minutes or chat history to extract case studies for similar situations, or aggregate each user's contribution or speaking tendency for use in talent evaluation or task allocation. Specifically, this server executes distributed relational databases and document-oriented databases. This server stores structured data in tables within the relational database. This server stores chat history or meeting minutes as unstructured data in the document-oriented database. This server uses natural language processing models to extract keyword vectors from text data. This server registers the keyword vectors as inverted indexes in the database. This server vectorizes the query upon receiving it. This server calculates the cosine similarity between vectors. This server extracts data in descending order of similarity. Thus, this system dramatically improves the retrieval speed for petabyte-scale data. This system eliminates storage I / O bottlenecks. This technology is applied to the knowledge management systems of multinational corporations.

[0069] The Security Management Department performs user authentication, access right management, communication encryption, data backup, information leakage countermeasures, unauthorized access detection, compliance with personal information protection regulations of various countries, audit log management, etc., to maintain a secure environment that multinational teams can use with peace of mind. Specifically, this server executes an authentication module based on a zero-trust architecture. This server verifies biometric data or hardware token information transmitted from a user terminal. This server applies a Transport Layer Security protocol to the entire communication path. This server executes a machine learning model for anomaly detection. This server converts access logs into time-series tensors. This server inputs the time-series tensors into an autoencoder network. This network outputs tensors that reconstruct the input data. This server calculates the reconstruction error between the input tensor and the output tensor. This server determines that an unauthorized access is made when the reconstruction error exceeds a threshold. Accordingly, the present system improves the defense capability against unknown cyberattacks. The present system reduces the labor cost required for security monitoring. The present technology is applied to international information sharing platforms of government agencies or financial institutions.

[0070] The Generative AI Collaboration Department receives requests from departments such as the Virtual Space Generation Department, the Communication Support Department, and the Project Schedule Management Department, generates appropriate prompts and sends them to the generative AI, and feeds the response results from the AI back to each department. For example, prompts such as "Please translate the Japanese statement 'ご検討のほどよろしくお願いいたします (We look forward to your consideration)' into an informal English expression adapted to American culture", "Please convert honorific expressions used in business meetings in Chinese culture into direct expressions adapted to German culture", "Please analyze the project schedule data and meeting logs of the past 3 months, identify factors causing schedule delays, and propose 3 improvement strategies", "Please summarize the minutes of this meeting and extract action items that take into account the cultural backgrounds of various countries", "Please detect signs of cultural friction from the chat history between specific members and propose mitigation strategies", and "Please generate a virtual office space for a multinational team with 3 sites in Asia, Europe, and North America, with a layout that reflects the cultural characteristics of each site" are sent to the generative AI. Specifically, this server executes a prompt engineering module. This server receives request data from each department as structured data in JSON format. This server parses the structured data and extracts necessary parameters. This server embeds the extracted parameters into a predefined prompt template. This server sends the generated prompt string to the API endpoint of a large language model. This server receives a text stream generated from the large language model. This server analyzes the received text and converts it into a data format for subsequent processing. This server asynchronously replies the converted data to the departments that are the request sources. Accordingly, the present system standardizes the interfaces between multiple modules and the generative AI. The present system minimizes the overhead of API calls. The present technology is applied to a microservice architecture that integrates multiple AI services.

[0071] Through such use cases, the system of this invention enables multinational and cross-cultural teams to collaborate smoothly and efficiently across physical, linguistic, and cultural barriers. It can be applied to various business operations of multinational and cross-cultural teams worldwide, including new business development, international R&D, remote-working startups, large corporations with multinational offices, international freelance communities, international conferences or academic societies, global education projects, international volunteer activities, cross-border supply chain management, international marketing activities, overseas office launch projects, international legal and intellectual property teams, and global customer support centers. Specifically, this system features a scalable server cluster built on a cloud computing environment. It uses a load balancer to distribute access requests from around the world to the optimal computing nodes. It uses container orchestration technology to execute each functional module as an independent container. It dynamically increases or decreases the number of container replicas based on traffic fluctuations. It leverages edge computing technology to perform data preprocessing closer to the user. Therefore, the system maintains stable service delivery even under sudden surges in access. The system maintains global system availability above 99.99%. This technology can be applied to metaverse platforms or international online event systems deployed on a global scale.

[0072] (Application Example 1) The specific processing flow corresponding to Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0073] (Implementation methods for carrying out the present invention) As an embodiment of the present invention, through the collaborative operation of servers and terminals, a highly specific and diverse range of application scenarios are envisioned to realize a virtual store operation support system supporting cross-border and cross-cultural operations. The following provides a comprehensive and thorough explanation of each component, the application method of generative AI, the role division between servers and terminals, and specific application examples and scenarios. Specifically, this system features a distributed database and a parallel processing cluster. The system establishes bidirectional real-time communication sessions with both customer and employee terminals. The system uses a 3D graphics engine to generate spatial data for the virtual store. The system utilizes a dedicated inference server cluster that hosts generative AI models. This inference server cluster is equipped with multiple graphics processing units to accelerate tensor operations. The system inputs customer action logs or purchase history into a stream processing pipeline in real time. Thus, the system processes massive amounts of transaction data with low latency. This system improves customer experience while reducing the fixed costs required for physical store operations. This technology is applied to e-commerce platforms in the globally expanding retail industry.

[0074] The virtual store space generation is implemented on the server side. The server receives various information from customers' and employees' terminals, including language, cultural background, time zone, purchase history, products they are interested in, virtual avatar settings, desired service style, age group, gender, past feedback, purchase frequency, time of day they visit the store, preferred payment method, type of terminal used, as well as interest information linked through SNS, past browsing history, friend relationships, and regional trend information, etc., as attribute information. Based on this information, a virtual store space optimized for each user is generated. For example, when customers from multiple regions such as Asia, Europe, North America, South America, Africa, and Oceania visit the store simultaneously, the system also acquires external data such as regional cultural characteristics or holidays, local trends, seasons, laws and regulations, logistics status, exchange rate information, popular product rankings, local event information, region-limited events, local currency display, local language price display, local delivery options, and local payment methods (electronic payment services, local bank transfers, virtual currencies, etc.) to dynamically optimize the layout, product display, promotional displays, reception style, event performances, product descriptions, background music, lighting, virtual avatar costumes, store decorations, digital signage, interactive displays, experiential demonstration booths, regionally categorized popular product zones, limited-edition product areas, VIP customer lounges, special areas for repeat customers, children's spaces for families, and accessible areas. For instance, during peak periods with many customers in the Asian region, the system automatically sets up decorations or background music that considers Asian culture and spatial performances that match local holidays or events; when there are many customers in the European region, it implements product displays or promotions that match European trends or seasons. Furthermore, based on customers' purchase history or preferred products, the system can automatically generate personalized product recommendation areas, special areas for repeat customers, VIP customer lounges, group shopping areas for friends, gift suggestion areas, high-rated product areas, sustainable product sections, and local producer introduction areas. Specifically, this server executes a spatial personalization algorithm. The server encodes customer attribute information or external data into multi-dimensional feature vectors. For example, a feature vector is a 256-dimensional array containing values ​​normalized to age groups or purchase frequencies. The server inputs the feature vectors into a deep reinforcement learning model. The model outputs an action value function representing the optimal store layout or product display pattern. For example, the output is an array of expected revenue scores for each layout candidate. The server selects the layout candidate with the highest score. Based on the selected layout, the server determines the configuration coordinates or texture IDs of 3D objects. The server serializes the determined spatial composition data and sends it to the customer's terminal. Thus, the system dynamically constructs a space that maximizes each customer's purchase intention. This system significantly improves conversion rates compared to static web pages. This technology is applied to personalized virtual shopping malls.

[0075] The communication support unit is mainly implemented on the server side, while part of the user interface and real-time display processing are also executed on the terminal side. The communication support unit uses generative AI to perform real-time multilingual translation on messages including text chats, voice calls, video receptions, whiteboard sharing, file sharing, emotional expressions via stickers or emojis, live broadcasts, interactive Q&A, product comparison chats, and gesture communication between avatars between customers and employees. For example, the question "What material is this product made of?" stated in Japanese is translated into English for English-speaking employees, and Japanese-style polite expressions are adjusted to direct expressions suited to British culture. Conversely, when a customer in the United States asks "What is the origin of this product?", the American informal inquiry style is converted to a formal expression suited to British culture while matching the English language of the employee. In addition, the generative AI refers to each user's cultural background, past communication history, purchase tendencies of various countries, religious considerations, gender expressions, selective use of honorific and informal expressions, meanings of non-verbal expressions, local buzzwords, dialects, slangs, industry terms, technical terms, abbreviations, product proper nouns, brand stories, storytelling techniques, local advertising expressions, local consumer protection regulations, local ethical norms, local social values, etc. For example, it converts expressions to reflect religious considerations for customers in Islamic cultural circles, automatically adjusts descriptions to emphasize quality or safety for customers in Chinese cultural circles, or converts expressions to emphasize sustainability or ethical consumption for customers in European cultural circles. Employees receive questions optimized for their native language and local culture, and generate appropriate responses with the support of AI while referring to product descriptions in the corresponding language from the product description database. For example, when an English-speaking employee answers "This is made from traditional Japanese materials.", the AI translates it into Japanese as "This is made of traditional Japanese materials", and automatically adds polite expressions or honorifics in accordance with Japanese culture. For Chinese customers, it is translated into Chinese, and expressions or emphasis points preferred in Chinese culture (such as emphasis on quality or safety, integration of tradition and modernity, care for families, response to gift-giving culture) are automatically added. In addition, the AI automatically generates cross-sell or upsell proposals at the timing when the customer's purchase intention increases, and can also display personalized promotional messages in real time such as "This is also recommended for people who purchased this product", "There is a discount for package purchase now", "Gift wrapping is also available", and "Earn points by posting reviews". Specifically, this server executes a multilingual-supporting Transformer model. This server splits the text message sent from the customer terminal into an array of subword tokens. This server generates a context vector representing the customer's cultural background or past purchase tendency.This server combines a labeled array and a context vector and inputs them into the model. The model outputs a labeled probability distribution for the target language. For example, the output is a time series of probability vectors with dimensions equal to the vocabulary size. This server uses a beam search algorithm to generate the optimal translated text. The server sends the generated text to employee terminals. The server also performs the same reverse translation process on responses from employees. Thus, this system not only prevents language barriers but also communication breakdowns caused by subtle cultural differences. The system minimizes the computational latency required for translation processing to milliseconds. This technology is applied to the real-time reception system of a cross-border e-commerce website.

[0076] Furthermore, using generative AI-based emotion recognition algorithms, the system analyzes customers' emotional states from multiple dimensions, including tone of voice, facial expressions, gestures, intonation, typing speed, frequency of emoji or sticker usage, avatar actions, changes in purchasing behavior (adding to cart or deleting items), comment content, feedback, sentiment analysis through SNS integration, and comparison with past purchasing experiences. This analysis is then visualized on the terminal screen through icons or colors, avatar facial expressions, emotion maps, heatmaps, real-time emotion scores, purchase intention indicators, satisfaction gauges, repurchase intention scores, and churn risk alerts. For example, a "Satisfied" icon is displayed when a customer is satisfied with the product description, while "Question" or "Dissatisfied" icons are displayed when there are questions or dissatisfaction. Employees can use these icons to provide additional explanations or follow-ups, apologize, offer special benefits, request questionnaires, request comments, or send messages encouraging repeat visits. Furthermore, the AI ​​automatically generates cross-selling or upsell proposals when customers' purchase intentions increase, and can also display personalized promotional messages in real time such as "Recommend this to people who bought this product," "Discounts available for package purchases now," "Gift wrapping also available," and "Earn points for leaving a comment." Specifically, this server executes a multimodal emotion recognition neural network. This server extracts facial feature point coordinates from camera images acquired from customer terminals. This server extracts acoustic features such as pitch or energy from speech data. This server calculates emotion polarity scores from text data. This server integrates these different modalities of data to generate a feature tensor. For example, the feature tensor is a three-dimensional floating-point array with a time axis. This network processes the feature tensor and outputs classification probabilities of emotion states. For example, the output is a probability vector of 0.85 for satisfaction and 0.05 for dissatisfaction. This server generates a display command for an emotion icon based on the probability vector. This server sends the display command to the employee terminal. Thus, this system accurately grasps the psychological state of customers during non-face-to-face interactions. This system improves the quality of employee responses and maximizes customer satisfaction. This technology is applied to online concierge services.

[0077] The purchase behavior analysis department is primarily implemented on the server side. The server accumulates the customer's action history within the virtual space (product browsing history, chat history, purchase history, feedback history, questionnaire results, rating scores, visit frequency, dwell time, cart addition history, checkout methods, return history, comment content, SNS sharing history, friend invitation history, activity participation history, coupon usage history, promotional response history, employee response history, popular product trends, seasonal sales fluctuations, regional purchasing tendencies, local economic indicators, local weather information, local traffic information, local logistics conditions, changes in local laws and regulations, etc.) as long-term memory, along with past promotional effects, employee response history, popular product trends, seasonal sales fluctuations, and regional purchasing tendencies. Generative AI is then used to perform multi-dimensional and time-series analysis on this massive amount of data. For example, if questions or purchases about a specific product are prevalent in a particular cultural sphere, the system automatically generates specific proposals such as, "Because this product is popular in Asia, we recommend strengthening promotions targeting Asian culture." In addition, suggestions for improving customer service quality are derived from employee response history. For example, "For English-speaking customers, using more informal language will increase satisfaction," "For Chinese-speaking customers, emphasizing quality assurance will increase purchase rates," and "For European customers, adding sustainability information will increase repeat purchase rates." Furthermore, AI, based on customer purchase behavior or feedback, extracts popular products, analyzes purchasing tendencies across different cultural regions, proposes optimizations for customer service quality, automatically generates promotional plans, provides employee training program prompts, designs special offers for returning customers, identifies VIP customers and provides special response plans, develops strategies to encourage repeat visits from customers at risk of churn, addresses shopping cart abandonment, prevents returns, promotes reviews, leverages social media for wider dissemination, designs friend invitation campaigns, promotes event participation, coordinates promotions with local events, addresses customer responses to local logistics delays, and adjusts sales strategies in response to changes in local laws and regulations. For example, it can extract products of interest to customers in the Asian region and their reasons from chat history over the past month, and propose three promotional plans. Alternatively, it can analyze high-satisfaction expression patterns from employee responses to customer questions and apply them to employee training. Specifically, this server executes time-series forecasting models and clustering algorithms. It converts customer action history or external factor data into time-series multidimensional tensors. These multidimensional tensors are then fed into a Long Short-Term Memory (LSTM) network. The network outputs a predicted score representing the probability of future purchases or churn risk. For example, the output might be a churn probability value for each customer in the range of 0 to 1. Based on these predicted scores, the server categorizes customers into multiple segments. For each segment, the server determines the optimal promotional strategy. The server registers the determined strategies in the promotional distribution engine. Finally, the server uses a natural language processing model to analyze employee response history and calculate a service skills evaluation score.Therefore, this system automates data-driven marketing strategies. It significantly reduces the analytical time required by marketing managers. This technology is applied to large-scale customer relationship management systems.

[0078] On the terminal side, the system is responsible for displaying the user interface, receiving user input, displaying real-time translation or sentiment analysis results, operating virtual avatars within the virtual space, browsing products, managing shopping carts, processing checkout, posting comments, receiving feedback, approving and rejecting AI proposals, inviting friends, sharing on social media, obtaining coupons, participating in activities, answering questionnaires, watching promotional videos, participating in live streams, interactive displays, virtual try-on, virtual tasting, virtual experiential demonstrations, switching local currency displays, selecting local delivery options, selecting local payment methods, accessibility settings, children's mode settings, and privacy mode settings. For example, when a customer enters a virtual store from the terminal, the server sends optimized store layout information and product lists, translated and culturally adapted product descriptions, visualizations of emotional states, promotional information, messages from employees, purchase history, privilege information, friends' store visits, event information, review rankings, social media sharing buttons, coupon obtaining buttons, questionnaire answering buttons, live stream participation buttons, and virtual try-on buttons to the terminal, which displays this content in real time. When a customer asks a question about a product, the terminal sends the content to the server. On the server side, generative AI performs translation, cultural adaptation, and sentiment analysis before distributing the optimized message to employee terminals. When an employee answers, their response is also optimized for the customer's terminal format. When a customer writes notes on a whiteboard, the content is automatically multilingualized, allowing employees from different countries to understand in their native languages. Specifically, this terminal utilizes its local graphics processing unit to perform 3D rendering. It maps store layout information and product lists received from the server into in-memory data structures. The terminal updates the camera matrix based on the user's viewpoint movement. It uses the camera matrix to project 3D objects onto a 2D screen. The terminal overlays translated product description text onto the UI layer. The terminal captures user clicks or swipes through event listeners. It converts the captured events into server-side request payloads. The terminal sends these payloads to the server via secure WebSocket communication. This completely offloads the rendering load from the server side, providing users with a smooth, lag-free experience. This technology is applied to virtual reality shopping applications for smartphones.

[0079] Furthermore, generative AI also learns user behavior patterns and purchasing tendencies, feedback, external trend information, national holidays and events, laws and regulations, logistics conditions, exchange rate fluctuations, local weather, local traffic, local social situations, local consumer protection regulations, local advertising regulations, local ethical norms, local social values, the activities of local consumer groups, the activities of local competitors, the activities of local influencers, local SNS trends, reviews on local word-of-mouth websites, reviews on local review websites, topics on local news websites, search trends on local search engines, popular videos on local video distribution services, popular songs on local music distribution services, popular books on local e-book services, and popular games on local game distribution services, etc., and contributes to the optimization of the entire system. For example, it can learn the tendency to become more active in purchasing during specific time periods, automatically propose important promotions or event schedules by AI, automatically adjust promotional content based on national holidays or large-scale events, predict the risk of exchange rate fluctuations or logistics delays to prompt sales strategies, automatically adjust sales strategies or product descriptions based on changes in local social situations or laws and regulations, or optimize promotional strategies by analyzing the activities of local competitors or the content posted by influencers. Furthermore, the AI ​​collects feedback and evaluations from customers and employees, continuously learning and improving translation accuracy, cultural adaptation processing, sentiment recognition algorithms, purchase behavior analysis algorithms, promotion optimization algorithms, employee training algorithms, repeat customer cultivation algorithms, VIP customer extraction algorithms, churn risk prediction algorithms, shopping cart abandonment countermeasures algorithms, return prevention algorithms, review promotion algorithms, SNS dissemination promotion algorithms, friend invitation promotion algorithms, event participation promotion algorithms, local event-linked promotion algorithms, customer response algorithms for local logistics delays, and sales strategy adjustment algorithms when local laws and regulations change. For example, when customers input feedback such as "This translation is unnatural," "This product description is difficult to convey its appeal in the local culture," "This AI proposal does not fit the actual situation," "This promotion does not conform to local trends," or "This employee's response violates local etiquette," the AI ​​learns the content and reflects it in future translations, proposals, promotions, employee training, and store operations. Specifically, this server executes a continuous learning pipeline. This server periodically crawls trend information and legal and regulatory data through external APIs. This server analyzes the collected text data using natural language processing models to construct a knowledge graph. This server appends relational vectors extracted from the knowledge graph to the micro-call dataset for generative AI. It inputs feedback text from customers or employees as reward signals into the reinforcement learning algorithm. The server updates the model's parameters using a proximal policy optimization algorithm. The updated model is tested and evaluated in a shadow environment. The server seamlessly replaces the production model when the evaluation score meets the benchmark.Therefore, this system autonomously maintains the model's adaptability to environmental changes. This system prevents model obsolescence and ensures long-term inference accuracy. This technology is applied to a self-evolving automated store operation platform.

[0080] In this embodiment of the invention, the functions of the virtual store space generation unit, communication support unit, and purchase behavior analysis unit are mainly implemented on the server side. On the terminal side, it is responsible for user interface and real-time display processing, user input reception, local optimization processing, promotion display, activity participation, comment posting, SNS sharing, friend invitation, coupon acquisition, questionnaire answering, live broadcast participation, virtual try-on, virtual experience demonstration, local currency display switching, local delivery option selection, local payment method selection, accessibility settings, children's mode settings, privacy mode settings, etc., thereby providing a highly supportive virtual store operation environment with generative AI at its core that supports cross-border and cross-cultural operations. For example, in all virtual stores supporting cross-border and cross-cultural collaborations, such as virtual stores of global fashion brands, online showrooms of international home appliance manufacturers, virtual tasting events of multinational food companies, virtual exhibitions targeting customers worldwide, cross-border e-commerce platforms, virtual textbook sales by international educational institutions, virtual auctions of global art galleries, virtual experience services of international tourist information centers, sales of local traditional handicrafts, direct sales of local agricultural products, sales of local tourist souvenirs, sales of works by local artists, sales of limited-edition products linked to local events, sales of local social contribution products, special editions of local sustainable products, special editions of local fair trade products, sales linked to local regional revitalization projects, sales linked to local crowdfunding, sales through local community collaborations, sales of textbooks in collaboration with local educational institutions, sales of health products in collaboration with local medical institutions, and sales of specialty products in collaboration with local autonomous regions, efficient and smooth reception and purchasing experiences can be achieved, overcoming physical, linguistic, cultural, social, economic, and legal barriers. Furthermore, through AI-driven automatic learning and optimization, the system continuously evolves, enabling it to flexibly respond to diverse user needs and changes in the global environment, local society, local economy, local laws and regulations, local trends, local consumer values, local competition dynamics, local influencer trends, local social media trends, local reputation website ratings, local review website ratings, local news website topics, local search engine trends, popular videos on local video distribution services, popular songs on local music distribution services, popular books on local e-book services, and popular games on local game distribution services. Specifically, this system adopts a microservice architecture, deploying each function as an independent service group. It uses service mesh technology to control and monitor communication traffic between services. The system executes edge AI models on the endpoint, completing some inference processing locally. It performs large-scale batch processing and real-time stream processing in parallel on the server side. It periodically transforms and loads unstructured data accumulated in the data lake into a data warehouse. The system collaborates with business intelligence tools to generate operational dashboards.Therefore, this system maximizes the fault tolerance and scalability of the entire system. This system meets the complex data governance requirements accompanying cross-border expansion. This technology is applied to the next generation of global metaverse business infrastructure.

[0081] (System Composition) The system involved in this implementation includes a virtual store space generation department, a communication support department, a purchase behavior analysis department, a user interface department, a database management department, a security management department, and an AI collaboration generation department. The virtual store space generation department receives various information from customers' or employees' terminals, such as language used, cultural background, time zone, purchase history, products followed, virtual avatar settings, desired service style, age group, gender, past feedback, purchase frequency, time of visit, preferred payment method, type of terminal used, interests and hobbies through SNS collaboration, past browsing history, friend relationships, and regional trend information. Based on this information, it generates a virtual store space optimized for each user. For example, when customers from multiple regions such as Asia, Europe, North America, South America, Africa, and Oceania visit the store simultaneously, the system also acquires external data such as regional cultural characteristics, holidays, local trends, seasons, laws and regulations, logistics status, exchange rate information, popular product rankings, local event information, region-specific events, local currency display, prices displayed in the local language, local delivery options, and local payment methods. This data is used to dynamically optimize the layout and product display within the virtual space, promotional displays, customer service style, event performances, product descriptions, background music (BGM), lighting, virtual avatar costumes, store decorations, digital signage, interactive displays, experiential demonstration booths, regionally categorized popular product zones, limited-edition product areas, VIP customer lounges, privileged areas for repeat customers, children's spaces for families, and accessible areas. For instance, during peak periods with many customers in the Asian region, the system automatically sets up decorations or BGM that consider Asian culture and spatial performances that match local holidays or events; when there are many customers in the European region, it implements product displays or promotions that align with European trends or seasons. Furthermore, it can automatically generate personalized product recommendation areas, privileged areas for repeat customers, VIP customer lounges, group shopping areas for friends, gift suggestion areas, highly rated product areas, sustainable product specials, and local producer introduction areas based on customers' purchase history or product preferences. Specifically, this virtual store space generation unit executes on a parallel computing cluster consisting of multiple computing nodes. This unit maps various attribute information received from customer terminals into a high-dimensional vector space. It uses a natural language processing model to convert text information into word embedding vectors. It normalizes numerical data such as purchase history or visit time periods to generate numerical vectors. This unit combines these vectors to construct an input tensor. For example, this input tensor contains a one-dimensional array representing customer age groups, a time-series array representing past purchase history, and word embedding vectors representing categorical variables related to cultural background. This input tensor is then fed into a graph neural network. This graph neural network learns a graph structure using nodes and edges to represent friend relationships between customers or regional trend information.This neural network updates the feature values ​​of each node and outputs space optimization parameters for each customer. The output data includes a three-dimensional floating-point tensor representing the layout coordinates within the virtual space, and a set of scalar values ​​representing lighting brightness or background music volume. Based on this output data, the virtual store space generation unit sends rendering commands to the rendering engine. Through this processing, the system eliminates layout changes made manually. The system dynamically generates the space based on customer attributes in milliseconds. As a result of this technology, the system extends customer dwell time and increases purchase intent by providing personalized spaces. In a specific application area, this system is used in virtual flagship stores of multinational apparel brands to automate product displays tailored to the culture of each region.

[0082] The Communications Support Department uses generative AI to perform real-time multilingual translation of messages between customers and employees, including text chats, voice calls, video calls, whiteboard sharing, file sharing, emotional expressions via stickers or emoticons, live streams, interactive Q&A, product comparison chats, and gesture communication between virtual avatars. For example, a question spoken in Japanese, "What material is this product made of?", is translated to an English-speaking employee as "What material is this product made of?", with the polite Japanese expression adjusted to a more direct British expression. Conversely, when a customer in the US asks, "What is the origin of this product?", the informal American question is translated into a more formal British expression while complementing the employee's English. Furthermore, the AI-generated content considers various factors, including users' cultural background, past communication history, purchasing preferences in different countries, religious considerations, gender expressions, distinctions between honorific and informal expressions, the meaning of nonverbal expressions, local slang, dialects, idioms, industry terms, technical terms, abbreviations, product-specific terms, brand stories, storytelling techniques, local advertising expressions, local consumer protection regulations, local ethical norms, and local social values. For example, it translates expressions reflecting religious considerations for customers from Islamic cultural backgrounds, automatically adjusts descriptions emphasizing quality or safety for customers from Chinese cultural backgrounds, and transforms expressions emphasizing sustainability or ethical consumption for customers from European cultural backgrounds. Employees receive questions optimized for their native language and culture. While referencing product descriptions in the corresponding language from a product description database, the AI ​​generates appropriate answers. For instance, if an English-speaking employee answers "This is made from traditional Japanese materials," the AI ​​translates it into Japanese, automatically adding polite expressions or honorifics to reflect Japanese culture. If the customer is from China, the content is translated into Chinese, automatically incorporating expressions or emphases favored in Chinese culture. In addition, AI can automatically generate cross-selling or upsell proposals when customers' purchase intentions increase, and can also display personalized promotional messages in real time such as "This is also recommended for people who have bought this product", "There is a discount for package purchases now", "Gift wrapping is also available", and "Review to earn points".Furthermore, using generative AI-based sentiment recognition algorithms, the system analyzes customers' emotional states from multiple dimensions, including tone of voice, facial expressions, gestures, intonation, typing speed, frequency of emoji or sticker usage, avatar actions, changes in purchasing behavior, frequency of adding or deleting items from shopping carts, comment content, feedback, sentiment analysis through SNS collaboration, and comparisons with past purchasing experiences. This analysis is then visualized on the terminal screen using icons or colors, avatar facial expressions, sentiment maps, heatmaps, real-time sentiment scores, purchase intention indicators, satisfaction gauges, repeat purchase intention scores, and churn risk alerts. Employees can use this information to provide additional explanations or follow-ups, apologies, offer privileges, request questionnaires, request comments, and encourage repeat visits. Specifically, this communications support department runs on an inference server hosting a large language model with a self-attention mechanism. The department segments customer input text into a tokenized sequence using a tokenizer. It generates context vectors representing the customer's cultural background or past purchasing tendencies. This communication support unit combines the labeled sequence and the context vector and inputs them into a large-scale language model. For example, the input data includes a labeled sequence of Japanese question text, a classification vector representing the customer's nationality, and a flag value representing past complaint history. This large-scale language model employs a Transformer architecture with billions of parameters. The large-scale language model calculates the dependencies between the elements of the input labeled sequence. The large-scale language model outputs a probability distribution of the labeled sequence in the target language. The output data includes string data representing the translated text and an array of floating-point numbers representing the selection confidence of each word. Furthermore, this communication support unit executes a multimodal emotion recognition model. This emotion recognition model receives an acoustic feature tensor containing the Mel-frequency cepstral coefficients of the customer's speech data and an image tensor representing the facial expression changes of a virtual avatar as input. This emotion recognition model uses a convolutional neural network to extract image features. This emotion recognition model uses a recurrent neural network to process the time-series acoustic features. This emotion recognition model outputs an emotion score vector representing the probability distribution of basic emotions such as joy, anger, and sadness. When the sentiment score vector exceeds a predetermined threshold, the communication support department sends a trigger signal to the subsequent promotion generation module. Through this process, the system does not simply translate literally, but performs real-time interpretive translation that takes cultural context into account. As a result, this system reduces communication errors between different cultures and improves customer satisfaction. In specific application areas, this system is used in customer support centers worldwide to handle inquiries from multilingual customers handled by a small number of specialized staff.

[0083] The purchasing behavior analysis department accumulates customers' action history in the virtual space (product browsing history, chat history, purchase history, feedback history, questionnaire results, rating scores, store visit frequency, dwell time, shopping cart entry history, checkout methods, return history, review content, SNS sharing history, friend invitation history, event participation history, coupon usage history, promotion response history, employee response history, popular product trends, seasonal sales fluctuations, regional purchasing tendencies, local economic indicators, local weather information, local traffic information, local logistics conditions, changes in local laws and regulations, etc.) as long-term memory, along with past promotional effects, employee response history, popular product trends, seasonal sales fluctuations, and regional purchasing tendencies. It then uses generative AI to perform multi-dimensional and time-series analysis on this massive amount of data. For example, when questions or purchases of a specific product are prevalent in a particular cultural sphere, the system automatically generates specific proposals such as, "Because this product is popular in Asia, we recommend strengthening promotions targeting Asian culture." In addition, suggestions for improving customer service quality are derived from employee interaction history. For example, "For English-speaking customers, using more informal language will increase satisfaction," "For Chinese-speaking customers, emphasizing quality assurance will increase purchase rates," and "For European customers, adding sustainability information will increase repeat purchase rates." Furthermore, AI, based on customer purchase behavior or feedback, performs tasks such as extracting popular products, analyzing purchase tendencies by cultural region, proposing optimizations for customer service quality, automatically generating promotional plans, providing employee training prompts, designing privileges for returning customers, extracting VIP customer information and providing special response plans, developing strategies to encourage repeat visits from at-risk customers, addressing shopping cart abandonment, preventing returns, promoting reviews, leveraging social media for promotion, designing friend invitation campaigns, encouraging participation in events, coordinating promotions with local events, handling customer issues related to local logistics delays, and adjusting sales strategies in response to changes in local laws and regulations. Specifically, this purchase behavior analysis department operates in collaboration with a distributed database system. This department constructs customer action history as a time-series tensor. This time series tensor contains timestamps of product browsing, dwell time in milliseconds, and a binary value indicating whether a shopping cart item was added. This purchase behavior analysis unit inputs this time series tensor into a Long Short-Term Memory (LSTM) network. The LSM network learns long-term dependencies in past behavior patterns. The LSM network outputs a score representing the probability of a customer making a future purchase. For example, this output data includes floating-point numbers from 0 to 1 representing the probability of purchasing a specific product category, and a percentage value representing the risk of churn. Based on this output score, the purchase behavior analysis unit clusters customers into multiple segments. The clustering results are then input into a reinforcement learning model. This reinforcement learning model explores the optimal promotional strategy based on a Markov decision process.This reinforcement learning model uses past promotional response history as a reward function. It maximizes the action value function representing the optimal discount rate or timing of a proposal for each customer segment. Based on the output of the reinforcement learning model, the purchase behavior analysis department generates specific promotional text. This process eliminates the rule-of-fact analysis performed by human marketers. The system automatically extracts non-trivial purchase patterns from high-dimensional action data. As a result, this system improves promotional conversion rates and prevents profit margin reductions caused by ineffective discounts. In specific application areas, this system is used in cross-border e-commerce platforms for dynamic pricing linked to seasonal changes or economic indicators in each region, and for the automated execution of promotional activities.

[0084] The User Interface (UI) department provides multilingual and multicultural interfaces, enabling customers or employees to operate virtual avatars in the virtual space. This allows for intuitive functions such as browsing products, managing shopping carts, processing checkout, posting reviews, inputting feedback, approving and rejecting AI proposals, inviting friends, sharing on social media, obtaining coupons, participating in events, answering questionnaires, watching promotional videos, participating in live streams, interactive displays, virtual try-ons, virtual tastings, virtual experiential demonstrations, switching local currency displays, selecting local delivery options, choosing local payment methods, accessibility settings, children's mode settings, and privacy mode settings. For example, when a customer enters a virtual store from their terminal, the server sends optimized store layout information and product lists, translated and culturally adapted product descriptions, visualizations of emotional states, promotional information, messages from employees, purchase history, privilege information, friends' store visits, event information, review rankings, social media sharing buttons, coupon obtaining buttons, questionnaire answering buttons, live stream participation buttons, and virtual try-on buttons to the terminal, which displays this content in real time. Specifically, this UI department utilizes the client terminal's graphics processing unit to perform rendering processing. This user interface unit parses the store layout information and the 3D coordinate data of the product list received from the server. This user interface unit transfers this data as vertex buffer objects to the graphics memory. This user interface unit uses vertex shaders and fragment shaders to render the polygonal model of the virtual space in real time. As a visualization of emotional state information, this user interface unit converts the emotional score vector received from the server into a color map. This user interface unit dynamically applies this color map to the texture of the virtual avatar. For example, as a purchase intention indicator, this user interface unit draws a progress bar with integer values ​​from 0 to 100 as its width. This user interface unit captures input operations from customers through event listeners. This user interface unit generates an event object containing the type of input operation and a timestamp. This user interface unit sends this event object to the server using an asynchronous communication protocol. Through this process, the system reduces the rendering load on the server side, achieving low-latency interface operations on the terminal side. As a result of this technology, the system provides a smooth virtual space experience even in environments with high network latency, preventing customer churn. In a specific application area, this system is used in mobile commerce applications that allow users with limited computing resources, such as smartphones or tablets, to experience high-quality virtual try-on or interactive displays.

[0085] The Database Management Department centrally manages and stores user information, virtual space composition information, communication history, purchase behavior data, product information, promotion history, feedback history, questionnaire results, evaluation scores, past project history, and various log data, enabling high-speed retrieval, extraction, and statistical analysis as needed. For example, it can retrieve past meeting minutes or chat history to extract case studies for similar situations, or statistically analyze each user's contribution or speaking tendencies for use in talent evaluation or task allocation, or efficiently provide large amounts of data for AI-driven progress analysis or risk prediction. Specifically, the Database Management Department employs a hybrid storage architecture integrating distributed key-value stores and vector databases. It stores structured data such as user information and purchase behavior data in a relational database. It uses natural language processing models to convert unstructured data such as chat history or product descriptions into high-dimensional word embedding vectors. The department indexes and stores these word embedding vectors in a vector database. For example, it converts the text of past meeting minutes into a 768-dimensional floating-point array. When a search query is input, it also converts the query into a word embedding vector. This database management department calculates the cosine similarity between the query vector and vectors within the database. It then rapidly extracts data with similarity exceeding a predetermined threshold. This extracted data is provided to the subsequent generative AI collaboration department. Through this process, the system extracts semantically similar past response cases that cannot be found through simple keyword consistency retrieval. As a result, this system reduces retrieval latency for large-scale datasets to milliseconds, enabling real-time information delivery. In a specific application area, this system is used in global retail companies with customer data from millions of individuals, serving as a knowledge management system to instantly extract optimal response patterns from massive past customer logs, supporting new employees' business operations.

[0086] The Security Management Department is responsible for user authentication, access control management, communication encryption, data backup, information leakage countermeasures, unauthorized access detection, compliance with personal information protection laws and regulations in various countries, and audit log management, maintaining a secure environment that multinational teams can use with confidence. For example, access permissions are configured in detail for each project, restricting access to specific confidential information or files; unauthorized access or malware attacks from external sources are detected and blocked in real time; and data storage destinations or data transmission paths are automatically optimized according to the laws and regulations of various countries. Specifically, the Security Management Department installs an access control mechanism based on a zero-trust network architecture. When a user attempts to log in, in addition to password authentication, the department processes multi-factor authentication requests containing biometric authentication data or the hardware fingerprint of the terminal. The department uses Advanced Encryption Standard (AES) algorithms to encrypt all data packets exchanged on the communication path. The department collects network traffic log data in real time and inputs it into an anomaly detection model. This anomaly detection model employs an autoencoder neural network. The model learns feature tensors of normal communication patterns. For example, the input data is a multi-dimensional numerical array containing the source IP address, port number, packet size, and communication frequency. This anomaly detection model compresses the input feature tensor into a low-dimensional latent space and then reconstructs it back to its original dimensions. The model calculates the reconstruction error between the input tensor and the reconstructed tensor. If this reconstruction error exceeds a dynamically set threshold, the security management department classifies the communication as unauthorized access. The department immediately terminates the session deemed unauthorized and sends an alert to the administrator. Through this process, the system autonomously defends against unknown cyberattacks that cannot be addressed by known signature-based detection methods. As a result, this system minimizes the risk of information leakage during global data transmission, ensuring compliance with the stringent data protection regulations of various countries. In specific application areas, this system is used as a secure data collaboration infrastructure across borders in cross-border e-commerce platforms that process customer credit card information or detailed purchase history.

[0087] The Generative AI Collaboration Department receives requests from various departments, including the Virtual Store Space Generation Department, Communication Support Department, and Purchase Behavior Analysis Department. It generates appropriate prompt text and sends it to the generative AI, then feeds back the AI's response to each department. Examples of prompts include: "Please translate the Japanese question 'What material is this product made of?' into a formal English expression reflecting British culture," "Please extract products that customers in the Asian region are interested in and their reasons from the past month's chat history, and propose three promotional plans," "Please analyze high-satisfaction expression patterns from employee responses to customer questions," "Please convert this product description into an expression emphasizing quality and safety for customers in the Chinese cultural sphere," "Please generate three incentive programs to encourage repeat purchases based on this customer's purchase behavior and feedback," "Please generate product descriptions to address changes in local laws and regulations," and "Please generate contingency plans to maintain customer satisfaction in the event of local logistics delays." The prompt text is sent to the generative AI. Specifically, this generative AI linkage unit functions as an API gateway in the microservice architecture. This unit temporarily stores asynchronous request messages from various units in a message queue. Based on the priority and type of the request, this unit selects an appropriate prompt template. It inserts contextual data from the request into the variable portion of the prompt template, dynamically generating the final prompt string. This unit sends the generated prompt string as an HTTP request to the inference endpoint of the natural language processing model. For example, this unit constructs a text array containing the customer's chat history and a payload defined by a JSON Schema specifying the required output format as input data. The inference endpoint executes a large language model to generate response text to the prompt. This unit parses the response text received from the inference endpoint and verifies its conformance to the specified JSON Schema. Finally, it returns the verified response data to the original request source module via a callback function. Through this process, the system efficiently schedules multiple AI inference requests simultaneously from multiple modules. As a result of this technology, the system prevents the inference server's computing resources from becoming exhausted, maximizing the overall system's processing throughput. In a specific application area, this system is used in a backend system that processes personalized message generation for each customer without delay during large-scale virtual events accessed simultaneously by thousands of customers.

[0088] In this way, the system involved in this implementation achieves an efficient and smooth reception and purchase experience that overcomes physical, linguistic, cultural, social, economic, and legal barriers in global, multicultural virtual stores by organically linking the virtual store space generation department, communication support department, purchase behavior analysis department, user interface department, database management department, security management department, and generative AI linkage department. Specifically, this system executes the above-mentioned departments as containerized independent processes and uses orchestration tools for unified management on a cluster. This system synchronizes the 3D spatial data constructed by the virtual store space generation department and the translated text data generated by the communication support department at high speed via shared memory space. This system dynamically injects the customer's purchase probability score output by the purchase behavior analysis department as the context of prompt words through the generative AI linkage department. This system executes these linkage processes asynchronously based on an event-driven architecture. For example, this system sends an inference request to the purchase behavior analysis department based on the action event of a customer picking up a specific product in a virtual image, and constructs a series of pipelines for the communication support department to provide reception suggestions to employees based on the result. Through this process, the system forms a feedback loop where each functional module operates independently, using each other's output data as input. As a result of this technology, the system adapts immediately to even subtle changes in customer behavior, providing a consistent and high-end customer service experience. In a specific application area, this system is used in the Metaverse stores of a high-end brand aiming to surpass the service quality of physical stores, serving as a unified platform to achieve a high level of personalization, as if each customer had their own dedicated consultant.

[0089] (Implementation steps) Step 1: Registration of user attribute information and initialization of virtual store space Customers and employees input various attribute information from their terminals, including language, cultural background, time zone, purchase history, products they are interested in, virtual avatar settings, desired service style, age group, gender, past feedback, purchase frequency, time of day they visit the store, preferred payment method, type of terminal used, interests linked through social networking, past browsing history, friend relationships, and regional trend information, and send this information to the server. The server receives this information and initializes a virtual store space optimized for each user. For example, when customers from multiple regions such as Asia, Europe, and North America simultaneously enter the store, it also acquires external data such as regional cultural characteristics and holidays, local trends, seasons, laws and regulations, logistics status, exchange rate information, popular product rankings, local event information, region-specific events, local currency display, local language price display, local delivery options, and local payment methods to dynamically optimize the layout, product display, promotional displays, service style, event performances, product descriptions, background music, lighting, virtual avatar clothing, and store decorations within the virtual space. Specifically, the server receives a JSON-formatted payload containing attribute information sent from the customer's terminal. The server parses the attribute values ​​within the received payload and inputs them into the feature extraction module. This feature extraction module converts categorical variables into binary vectors using one-hot encoding. It then normalizes continuous variables using min-max scaling to generate floating-point vectors. The server combines these vectors to construct a customer-specific profile tensor. For example, this profile tensor contains a 5-dimensional binary array representing age groups and a 1-dimensional normalized value representing past purchase amounts. The server inputs this profile tensor into a spatial generative neural network. This network uses a multilayer perceptron to convert the profile tensor into a representation vector of the latent space. Based on this representation vector, the network outputs a set of transformation matrices representing the configuration coordinates, scaling ratios, and rotation angles of the objects constituting the virtual space. The server obtains environmental data vectors containing local weather and exchange rate information via an external API and combines them with the representation vectors to determine the final spatial layout. Through this process, the system does not select a static template but dynamically synthesizes spaces with countless variations based on the combination of customer attributes and the external environment. As a result, the system instantly provides a highly immersive virtual environment perfectly aligned with the customer's cultural background and current situation, dramatically increasing initial engagement. As a specific application area, this system is used to automatically construct customized exhibition booths that match the areas of interest and home culture of various visitors in international online exhibitions visited by customers around the world.

[0090] Step 2: Multilingual and Multicultural Communication Support Customers and employees utilize various communication methods in the virtual space, including text chat, voice calls, video conferencing, whiteboard sharing, file sharing, emotional expression via stickers or emoticons, live streaming, interactive Q&A, product comparison chat, and gesture communication between virtual avatars. The server employs generative AI to perform real-time multilingual translation of messages between customers and employees, automatically adapting the language based on factors such as each user's cultural background, past communication history, purchasing preferences from different countries, religious considerations, gender expressions, distinction between honorific and informal language, the meaning of nonverbal expressions, local slang, dialects, idioms, industry terms, technical terms, abbreviations, product terminology, brand stories, storytelling techniques, local advertising expressions, local consumer protection regulations, local ethical norms, and local social values. Examples of prompts for the generative AI include, "Please translate 'What material is this product made of?' spoken in Japanese into a formal English expression in British culture," and "Please convert this product description into an expression emphasizing quality and safety for customers in the Chinese cultural sphere." Specifically, this server inputs the voice data streamed from the customer's terminal into the speech recognition engine, converting it into a text token sequence. The server then combines this text token sequence with a cultural code vector extracted from the customer's profile tensor and inputs it into a Transformer model for translation. This Transformer model employs an encoder-decoder architecture. The encoder calculates the contextual meaning of the input text token sequence using a self-attention mechanism, generating a hidden state tensor. The decoder, referring to the hidden state tensor and the cultural code vector, autoregressively generates a text token sequence for the target language. For example, the input data includes a text string derived from Japanese speech and a multi-dimensional vector representing formal British business culture. The decoder's output data includes a translated English text string and a floating-point array representing the generation probability of each word. The server inputs the generated text token sequence into a speech synthesis engine to generate speech waveform data in the target language. The server then streams this speech waveform data to employee terminals in real time. Through this process, the system not only eliminates language barriers algorithmically but also eliminates misunderstandings caused by subtle cultural differences. As a result of this technology, the system facilitates smooth communication between employees and customers across borders, increasing the success rate of business negotiations. In specific applications, this system is used in virtual showrooms for selling high-value real estate or automobiles to overseas customers, serving as a multilingual translation system that supports detailed reception services that take local culture into account.

[0091] Step 3: Visualization of Emotion Recognition and Purchase Intent The server uses a generative AI-based emotion recognition algorithm to analyze customers' emotional states from multiple dimensions, including tone of voice, facial expressions, gestures, intonation, typing speed, frequency of emoji or sticker usage, avatar actions, changes in purchasing behavior (adding to cart or deleting items), comment content, feedback, sentiment analysis through SNS integration, and comparison with past purchasing experiences. This analysis is then visualized on the terminal screen using icons or colors, avatar facial expressions, emotion maps, heatmaps, real-time emotion scores, purchase intention indicators, satisfaction gauges, repeat purchase intention scores, and churn risk alerts. For example, a prompt to the generative AI could be, "Based on this customer's chat history and purchasing behavior, analyze their current emotional state and purchase intention, and generate a score for visualization." Specifically, the server executes a multimodal emotion recognition pipeline. The server synchronously includes strings of text chat received from the customer's terminal, acoustic features of voice data, and time-series data from the avatar's operation logs. The server inputs the text strings into a natural language processing model to extract emotion vectors representing emotional polarity. This server inputs voice data into a convolutional neural network to extract acoustic emotion vectors representing the intonation and rhythm of the voice. It calculates movement speed and gesture frequency from the virtual avatar's operation logs to generate behavioral feature vectors. These vectors are concatenated in a splicing layer to construct a comprehensive feature tensor. This comprehensive feature tensor is then input into a fully connected layer, outputting an emotion score vector representing the probability distribution of basic emotion categories such as joy, anger, sadness, and surprise. For example, the output data might contain floating-point arrays representing joy with a probability of 0.8 and anger with a probability of 0.05. The server smooths this emotion score vector sequentially to calculate a scalar value representing purchase intention. This scalar value is sent to the terminal's rendering engine and used as a display parameter for heatmaps or gauges on the screen. Through this processing, the system quantifies subtle nonverbal information that human employees might easily overlook and provides it as an objective indicator. As a result of this technology, the system detects early signs of customer dissatisfaction or churn, allowing employees to follow up at the appropriate time. As a specific application area, this system is used in face-to-face sales of high-value goods online, to monitor customers' psychological state in real time, and to notify employees of the best time to close the deal.

[0092] Step 4: AI-based promotion and reception optimization proposals The server uses generative AI to automatically generate cross-selling or upsell proposals when customers' purchase intentions increase, and displays "Recommend this product to customers who have purchased this product as well" in real time. Personalized promotional messages such as "Discounts available for package purchases now," "Gift wrapping also available," and "Earn points for posting reviews" are displayed. Additionally, analyzing employee response history and customer feedback automatically generates suggestions and training programs to improve customer service quality. Examples of prompts for the generative AI include, "Please extract products that customers in the Asian region are interested in and their reasons from the past month's chat history, and propose three promotional plans," and "Please analyze high-satisfaction expression patterns from employee responses to customer questions." Specifically, this server detects that the scalar value of purchase intention output from the sentiment recognition pipeline exceeds a predetermined threshold. This detection event triggers the recommendation system module. This module inputs the customer's past purchase history vector and the feature vector of the currently viewed product into a collaborative filtering algorithm. This collaborative filtering algorithm performs matrix factorization with data from other customers with similar purchasing tendencies, outputting a score list of recommended candidate products. The server extracts the top three product IDs with the highest scores. The server then inputs the extracted product IDs and the customer's cultural code vector into the generative AI model. This generative AI model generates a token sequence of promotional text that combines product features with the customer's cultural background. For example, the input data includes a JSON object containing metadata about the recommended product and an array of texts from past promotional articles that received positive customer feedback. The output data of this generative AI model includes a personalized proposal message string and a timestamp indicating the timing of the display. The server sends the generated proposal message to the user interface of the customer's terminal and displays it as a pop-up window. Through this process, the system does not execute a generic advertisement display, but rather a dynamic proposal that is completely consistent with the customer's current psychological state and context. As a result of this technology, the system statistically significantly improves the success rate of cross-selling and upselling, maximizing the average order value. In a specific application area, this system is used in virtual department stores to provide an AI-powered personalized advisor function that, at the moment a customer adds a specific item to their shopping cart, proposes matching accessories or related products in a natural conversational manner.

[0093] Step 5: Purchase Behavior Analysis and Store Operation Optimization The server accumulates customer behavior history within the virtual space (product browsing history, chat history, purchase history, feedback history, questionnaire results, rating scores, store visit frequency, dwell time, shopping cart history, checkout methods, return history, review content, SNS sharing history, friend invitation history, activity participation history, coupon usage history, promotion response history, employee response history, popular product trends, seasonal sales fluctuations, regional purchasing tendencies, local economic indicators, local weather information, local traffic information, local logistics conditions, changes in local laws and regulations, etc.) and past promotional effects, employee response history, popular product trends, seasonal sales fluctuations, and regional purchasing tendencies as long-term memory. Generative AI then performs multi-dimensional and time-series analysis on this massive amount of data. For example, prompts such as "Based on this month's purchase data and chat history, extract customers with high churn risk and propose three strategies to encourage repeat visits," "Generate product descriptions to address changes in local laws and regulations," and "Generate solutions to maintain customer satisfaction during local logistics delays" are sent to the generative AI. Specifically, this server periodically batch processes large-scale action history logs accumulated in the data warehouse. It extracts session data for each customer from these logs and shapes it into a time-series event sequence. This event sequence is then fed into an autoencoder-based anomaly detection model. This model learns patterns of normal purchasing behavior and calculates the reconstruction error of the input event sequence. Sessions with large reconstruction errors and no purchases are identified as high-risk churn patterns and extracted. The extracted patterns are then fed into a clustering algorithm, grouping them by each churn factor. Representative feature vectors from each group are input into a generative AI model. This model analyzes churn factors and generates text reports representing improvement strategies. For example, the input data includes a numerical array representing the time-series trend of shopping cart abandonment rates in a specific region, and text data from news reports about logistics delays in that region. The output data of the generative AI model includes templates for messages apologizing to customers for logistics delays and definition data for business processes proposing alternative delivery options. The generated response is then displayed on a dashboard for store managers. Through this process, the system discovers correlations in massive log data that are difficult for humans to perceive and transforms them into concrete action plans. As a result of this technology, the system automatically identifies bottlenecks in store operations and improves overall sales by supporting data-driven decision-making. In a specific application area, this system is used in e-commerce companies operating globally to support the development of business strategies that enable rapid responses to changes in market trends and external factors in various regions.

[0094] Step 6: Provide information and operations through the user interface On the terminal side, a multilingual and multicultural interface is provided, allowing users to operate virtual avatars in a virtual space and intuitively browse products, manage shopping carts, process checkouts, post comments, input feedback, approve and reject AI proposals, invite friends, share on social media, obtain coupons, participate in activities, answer questionnaires, watch promotional videos, participate in live streams, experience interactive displays, virtually try on clothes, virtually taste food, perform virtual demonstrations, switch local currency displays, select local delivery options, select local payment methods, accessibility settings, children's mode settings, and privacy mode settings. The terminal displays real-time information from the server, including store layout information and product lists, translated and culturally adapted product descriptions, visualizations of emotional states, promotional information, messages from employees, purchase history, discount information, friends' store visits, event information, review rankings, social media sharing buttons, coupon obtaining buttons, questionnaire answering buttons, live stream participation buttons, and virtual try-on buttons. Specifically, the terminal uses a built-in processor and graphics processing unit to process data received from the server. The terminal caches the 3D mesh data and texture data of the store layout sent from the server in local memory. This terminal samples input signals from the user's touchscreen operations or controllers, calculating the virtual avatar's movement and gaze vectors. Based on these vectors, the terminal updates the virtual camera's transformation matrix, re-rendering the virtual space scene at a rate of over 60 frames per second. In a separate thread from the main rendering loop, the terminal processes translated text messages and sentiment scores asynchronously received from the server. Based on the sentiment score data, the terminal dynamically changes the transparency and hue of user interface elements on the screen. For example, the terminal normalizes the purchase intention indicator value to the range of 0 to 1 and uses this value as an interpolation coefficient for a color gradient from red to green. When the terminal detects that the user has pressed the virtual try-on button, it activates the camera and extracts the user's skeletal feature points. The terminal synthesizes the extracted skeletal feature points and a 3D model of the product, overlaying them onto the augmented reality space. Through this processing, the system optimally distributes the computing resources of the server and terminal, providing a rich user experience while suppressing network bandwidth consumption. As a result of this technology, the system minimizes operational latency, enhancing immersion and interactivity within the virtual space. As a specific application area, this system is used in a next-generation online shopping application that allows consumers to have a realistic product confirmation and try-on experience, just like in a physical store, even when they are at home.

[0095] (specific use case) For example, consider a multinational brand with stores in Asia, Europe, and North America operating virtual stores for customers worldwide. The virtual store space generation department receives information from each customer, including their language, cultural background, time zone, purchase history, products they are interested in, avatar settings, desired service style, age group, gender, past feedback, purchase frequency, time of day, preferred payment method, type of device used, SNS-based interests, browsing history, friend relationships, and regional trends. For customers in Asia, it automatically generates decorations or background music matching local holidays or events, displays local currency, provides local delivery options, and displays layouts reflecting popular local products. For customers in Europe, it offers sustainable product sections or promotions reflecting local trends. For customers in North America, it provides family-friendly children's areas or limited-edition product areas linked to local events, and supports local electronic payment services, providing a virtual space optimized for each region. Specifically, this virtual store space generation department receives access requests from customers in each region via a load balancer. The department determines the geographical location from the IP address of the access source and sends API requests to the corresponding external data source. This virtual store space generation department parses the acquired local holiday information or popular product ranking data in JSON format. It combines customer attribute information vectors with the acquired external data vectors and inputs this data into the space layout generation model. This space layout generation model functions as a generator within a generative adversarial network (GAN). Based on the input vectors, the generator outputs three-dimensional voxel data of the virtual space. For example, the input data includes categorical variables representing customers from the Asian region and marker values ​​representing local Lunar New Year activities. The generator's output data includes configuration coordinate tensors for decorative objects with red and gold tones, and metadata indicating price display in local currency. The virtual store space generation department converts the generated voxel data into a polygon mesh and sends it to the customer's terminal. Through this process, the system dynamically generates completely independent customized spaces for each region on a single platform. As a result, this system provides customers in various regions with the most user-friendly shopping environment, strengthening brand loyalty in the global market. In a specific application area, this system is used in virtual new product launches held simultaneously worldwide, automatically switching the displayed content or performances based on the participants' residential regions—a global marketing platform.

[0096] The communication support unit, when a customer asks a question in Japanese "What material is this product?" (この商品はどんな素材ですか?), sends a prompt to the generative AI stating "Please translate the Japanese utterance 'What material is this product?' into a formal English expression adapted to British culture", and conveys the culturally adapted English "What material is this product made of?" to the staff. Conversely, when a customer in the Americas asks "What is the origin of this product?", the communication support unit sends a prompt to the generative AI stating "Please translate the American English utterance 'What is the origin of this product?' into a polite Japanese expression adapted to the Japanese cultural sphere", and conveys the polite Japanese "This product is manufactured in a traditional Japanese workshop" (こちらの商品は日本の伝統的な工房で製造されております) to the staff. In addition, the generative AI automatically performs cultural adaptation processing by referring to each user's cultural background, past communication history, purchasing tendencies of various countries, religious considerations, gender expression, differentiated use of honorific and informal expressions, non-verbal expressions, local buzzwords, dialects, slang, industry terms, professional terminology, abbreviations, product-specific proper nouns, brand stories, local advertising expressions, local consumer protection regulations, local ethical norms, local social values, and the like. Specifically, the communication support unit converts the customer's voice input into text, and determines the source language through the language determination module. The communication support unit loads the weight parameters of the appropriate translation model into the memory based on the determined pair of the source language and the target language. The communication support unit combines the token sequence of the input text with the instruction prompt constructed by the prompt generation module, and inputs the combined result into the large language model. This instruction prompt includes a control vector that specifies the directionality of translation, cultural tone, and honorific level. For example, the input data includes the Japanese character string "What material is this product?" and a metadata tag that specifies "formal English expression adapted to British culture". The large language model uses the self-attention mechanism of Transformer to analyze the contextual dependencies between input tokens. The large language model generates and outputs a token sequence of the target language. The output data includes the English character string "What material is this product made of?" and a score value indicating translation confidence. The communication support unit forwards the output character string to the chat interface of the staff terminal. Through this processing, the system does not perform simple dictionary-based replacement, but executes advanced semantic conversion that deeply understands the context and cultural background. As a technical effect of the present technology, the system eliminates cultural frictions between customers and staff, and promotes the establishment of trust relationships.As a specific application area, this system is used in the virtual concierge desk of high-end hotels targeting international customers, and is an automatic translation system that responds with appropriate wording that conforms to the etiquette and customs of various countries.

[0097] The purchasing behavior analysis department accumulates customer behavior history (product browsing history, chat history, purchase history, feedback history, questionnaire results, rating scores, store visit frequency, dwell time, add-to-cart history, checkout method, return history, review content, SNS sharing history, friend invitation history, event participation history, coupon usage history, promotion response history, employee response history, popular product trends, seasonal sales fluctuations, regional purchasing tendencies, local economic indicators, local weather information, local traffic information, local logistics conditions, local regulatory changes, etc.) as long-term memory. This data is then sent to the generative AI system with commands such as, "From the purchase data and chat history of the past month, extract customers with high churn risk and propose three strategies to encourage repeat visits," and "From the chat history of the past month, extract products that customers in the Asian region are interested in and the reasons why, and propose three strategies to encourage repeat visits." The system automatically performs tasks such as prompting keywords like "Promotional Plan," "Analyze high-satisfaction expression patterns from employee responses to customer inquiries," "Generate product descriptions to address local regulatory changes," and "Generate response plans to maintain customer satisfaction during local logistics delays." Based on AI's multidimensional and time-series analysis results, it automatically performs tasks including extracting popular products, analyzing purchasing tendencies across different cultural regions, proposing optimizations for customer service quality, automatically generating promotional plans, prompting employee training plans, designing privileges for returning customers, extracting VIP customers and prompting special response plans, developing strategies to encourage repeat visits from at-risk customers, implementing strategies to address shopping cart abandonment, preventing returns, promoting reviews, leveraging social media for wider dissemination, designing friend invitation campaigns, promoting event participation, co-promoting local events, developing customer response plans for local logistics delays, and adjusting sales strategies in response to local regulatory changes. Specifically, the purchasing behavior analysis department uses a parallel data processing framework to aggregate petabyte-level action history logs stored in a distributed file system. This aggregated data is then constructed into a multidimensional tensor and input into a time-series forecasting model. This time series forecasting model employs a hybrid architecture combining convolutional neural networks and long short-term memory networks. It simultaneously learns the spatial characteristics and temporal progression of past purchasing patterns. The purchasing behavior analysis unit combines the output of the time series forecasting model with prompts for generative AI and inputs them into a large language model. For example, the input data includes a time series array representing the sales trend of a specific product over the past month, and a natural language prompt string stating, "Identify customers at high risk of churn and propose three strategies to encourage repeat visits." This large language model comprehensively interprets the input numerical data and textual prompts to generate specific policy implementation plans. The output data is a structured JSON object containing the promotional title of the proposal, conditional expressions for target customer segmentation, and predicted values ​​for expected results. The purchasing behavior analysis unit parses this JSON object and sends it as an execution command to the marketing automation system.Through this process, the system automatically executes the entire process from data analysis to policy formulation and execution, end-to-end, without human intervention. As a result, the system dramatically improves the speed of response to market changes and minimizes opportunity loss. In a specific application area, this system is used in large-scale e-commerce websites handling tens of thousands of products, providing a self-managing store operation system that analyzes daily sales data and customer feedback in real time and automatically optimizes the next day's sales strategy.

[0098] The user interface department provides a multilingual and multicultural interface, enabling customers to operate virtual avatars in the virtual space and intuitively browse products, manage shopping carts, process checkouts, post reviews, input feedback, approve and reject AI proposals, invite friends, share on social media, obtain coupons, participate in activities, answer questionnaires, watch promotional videos, participate in live streams, experience interactive displays, virtually try on clothes, virtually sample food, engage in virtual experiential demonstrations, switch local currency displays, select local delivery options, select local payment methods, and accessibility settings, children's mode settings, and privacy mode settings. For example, when a customer enters a virtual store from their terminal, the server sends optimized store layout information and product lists, translated and culturally adapted product descriptions, visualizations of emotional states, promotional information, messages from employees, purchase history, privilege information, friends' store visits, event information, review rankings, social media sharing buttons, coupon obtaining buttons, questionnaire answering buttons, live stream participation buttons, and virtual try-on buttons to the terminal, which displays this content in real time. Specifically, this user interface department utilizes a WebGL-based rendering engine running on the customer's web browser. This user interface unit asynchronously receives updated data on store layout and product lists from the server via WebSocket communication. It stores the received data in a local state management library and updates the virtual DOM tree. The user interface unit handles input for movement or viewpoint changes of virtual avatars within the virtual space using quaternion rotation calculations. It displays visualized emotional states or promotional information received from the server as billboard objects in 3D space, following above the virtual avatar's head. For example, it parses the "purchase history" array data received from the server and dynamically maps it as textures for digital signs configured on the walls within the virtual space. When a customer clicks the virtual try-on button, the user interface unit executes a shader program that uses the terminal's GPU to synthesize a 3D model of the product and the customer's camera image in real time. Through this processing, the system efficiently handles complex 3D rendering and multilingual text rendering on the terminal side. As a result of this technology, the system provides customers with an intuitive and responsive user experience while distributing server load. Specifically, this system is used in cross-platform e-commerce services that provide a high-quality virtual shopping experience through a web browser without requiring the installation of a dedicated application.

[0099] In such use cases, specific examples of prompt texts for generative AI to read include: "Please translate 'What material is this product made of?' spoken in Japanese into a formal English expression in British culture," "Please translate 'What is the origin of this product?' spoken in American English into a polite Japanese expression in the Japanese cultural sphere," "Please extract customers with high churn risk from the past month's purchase data and chat history, and propose three strategies to encourage repeat visits," "Please extract products of interest to Asian customers and their reasons from the past month's chat history, and propose three promotional plans," "Please analyze high-satisfaction expression patterns from employee responses to customer questions," "Please generate product descriptions to address changes in local regulations," and "Please generate response plans to maintain customer satisfaction during local logistics delays," etc. Specifically, this system formats these prompt texts into a form suitable for the input interface of the generative AI model. The system uses a tokenizer to convert the prompt text into an array of integer indices. This array of indices is then input into the encoder of the Transformer model. To accurately convey the intent of the prompt text to the model, this system employs prompt engineering techniques, attaching system role definitions or constraints at the beginning of the input text. For example, the input data might be a text tensor containing the system prompt "You are an excellent cross-cultural communication assistant" and the user prompt "Please translate 'What material is this product made of?' spoken in Japanese into a formal English expression of British culture." The system dynamically adjusts the temperature parameter or Top-p sampling threshold during model output generation to control the balance between output diversity and accuracy. The system de-converts the text tag sequence output by the model into a string using a decoder. Through this processing, the system transforms ambiguous instructions given in natural language into specific, computer-executable tasks, leading to high-precision inference results. As a result, this system enables the flexible handling of diverse use cases using a single, general-purpose AI model, reducing system development costs and time. In a specific application area, this system is used in a no-code business support interface that automates complex data analysis or translation tasks even for store employees without programming knowledge, by allowing them to give instructions to the system in natural language.

[0100] In this way, the system of the present invention can provide diverse customers worldwide with virtual store experiences optimized for language, culture, social background, purchasing preferences, and local conditions, while maximizing global sales opportunities, improving customer satisfaction, and optimizing store operations. Specifically, this system functions as a distributed computing architecture integrating a virtual store space generation unit, a communication support unit, a purchase behavior analysis unit, and a user interface unit. The system exchanges high-dimensional tensor data or embedding vectors between modules in real time. The system continuously provides all customer input data or action logs to the inference pipeline of the machine learning model. Based on the inference results, the system dynamically updates the rendering parameters of the virtual space or the weights of the translation model. For example, the system sets customer dwell time or purchase conversion rate as the objective function and autonomously optimizes the parameters of the entire system using reinforcement learning algorithms. The system uses a GPU cluster on the cloud to process this optimization process in parallel. Through this process, the system completely eliminates static, rule-based processing, constructing a data-driven and self-evolving adaptive information processing foundation. As a result of this technology, the system performs personalization on a scale and speed exceeding the limits of human cognition, establishing a competitive advantage in the global market. As a specific application area, this system is used as the core system of a next-generation metaverse business platform that transcends national and linguistic barriers and provides services to millions of users simultaneously.

[0101] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0102] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0103] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0104] The data generation model 58 is obtained by performing deep learning on a neural network. A prompt containing instructions is input to the data generation model 58, along with inference data such as speech data representing speech, text data representing text, and image data representing images (e.g., data from still images or moving images). The data generation model 58 infers the input inference data based on the instructions shown in the prompt and outputs the inference result in one or more data forms, such as speech data, text data, and image data. The data generation model 58 may include, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization. The specific processing unit 290 performs the aforementioned specific processing while using the data generation model 58. The data generation model 58 may also be a fine-tuned model to output inference results from prompts without instructions; in this case, the data generation model 58 can output inference results from prompts without instructions. The data processing device 12, etc., includes multiple data generation models 58, including AI other than generative AI. AI other than generative AI includes, but is not limited to, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes. Furthermore, AI can also act as an AI agent. Additionally, when the processing described above is performed by AI, this processing may be partially or entirely performed by AI, but is not limited to this example. Moreover, processing performed by AI, including generative AI, can be replaced by rule-based processing, and vice versa.

[0105] Furthermore, the processing performed by the data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. In addition, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0106] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0107] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0108] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. One example of the data processing device 12 is a server.

[0109] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0110] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0111] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0112] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0113] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0114] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0115] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0116] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0117] In the smart glasses 214, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290.

[0118] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0119] (Example 1) Since the specific processing flow is the same as that in Embodiment 1 described in the first embodiment above, the description is omitted.

[0120] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0121] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0122] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0123] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0124] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0125] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. An example of the data processing device 12 is a server.

[0126] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0127] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0128] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0129] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0130] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0131] Figure 6 An example of the main functions of the data processing device 12 and the head-mounted terminal 314 is shown. Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0132] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0133] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0134] In the head-mounted terminal 314, the processor 46 performs the receiving and output processing.

[0135] The memory 50 stores a receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0136] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0137] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0138] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0139] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0140] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the head-mounted terminal 314, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0141] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the head-mounted terminal 314.

[0142] [Fourth Implementation] Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0143] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0144] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0145] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control object 443. Computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, control object 443, and communication I / F 44 are also connected to the bus 52.

[0146] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0147] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0148] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0149] The controlled object 443 includes a display device, LEDs for the eyes, and motors for driving the arms, hands, and feet. The posture and movements of the robot 414 are controlled by controlling the motors for the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, facial expressions of the robot 414 can also be expressed by controlling the illumination state of the LEDs for the robot 414's eyes.

[0150] Figure 8 An example of the main functions of the data processing device 12 and the robot 414 is shown. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0151] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0152] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0153] In robot 414, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on RAM 48.

[0154] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0155] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0156] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0157] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0158] Furthermore, the processing performed by the data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0159] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the robot 414.

[0160] Furthermore, the emotion-specific model 59, serving as an emotion engine, can determine the user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine the user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The system determines the user's emotions. In addition, the emotion-specific model 59 can also determine the robot's emotions in the same way, and the specific processing unit 290 can also perform specific processing using the robot's emotions.

[0161] Figure 9 This is a diagram representing an emotion map 400 that maps various emotions. In the emotion map 400, emotions are arranged radially from the center in concentric circles. The closer to the center of the concentric circles, the more primitive the emotion is. Further out on the concentric circles, emotions are arranged representing states or actions arising from mood. Emotion is a concept that includes both feelings and mental states. To the left of the concentric circles, emotions generated by reactions occurring in the brain are arranged roughly. To the right of the concentric circles, emotions guided by situational judgments are arranged roughly. Above and below the concentric circles, emotions generated by reactions occurring in the brain and guided by situational judgments are arranged roughly. Furthermore, the emotion of "pleasure" is arranged above the concentric circles, and the emotion of "unpleasantness" is arranged below. Thus, in the emotion map 400, various emotions are mapped according to the structure of emotion generation, while easily generated emotions are mapped nearby.

[0162] These emotions are distributed at the 3 o'clock position on the Emotion Chart 400, and usually fluctuate between peace and unease. In the right half of the Emotion Chart 400, because situational awareness is more dominant than internal feelings, it gives a sense of calm.

[0163] The inner side of the emotion diagram 400 represents the mind, and the outer side of the emotion diagram 400 represents actions. Therefore, the further you go to the outer side of the emotion diagram 400, the more the emotion can be seen (manifested in actions).

[0164] Here, human emotions are based on a balance of various factors such as posture and blood sugar levels. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. In robots, cars, and motorcycles, emotions can also be created based on a balance of factors such as posture and remaining battery power. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Speech Emotion Recognition and Brain Physiological Signal Analysis Systems for Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the "Reaction" domain, where sensation is dominant, are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the "Situation" domain, where situational cognition is dominant, are arranged.

[0165] The emotion map defines two types of emotions that promote learning. One is a negative emotion located near the middle of "repentance" or "reflection" on the situation side. That is, when the robot experiences negative emotions such as "I never want to feel this way again" or "I never want to be scolded again." The other is a positive emotion located near "desire" on the response side. That is, when the robot experiences positive feelings such as "wanting more" or "wanting to know more."

[0166] The emotion-specific model 59 feeds user input into a pre-learned neural network to obtain emotion values ​​representing each emotion shown in the emotion graph 400, and determines the user's emotion. This neural network is pre-learned based on multiple learning data sets that combine user input with emotion values ​​representing each emotion shown in the emotion graph 400. Furthermore, this neural network is learned to... Figure 10 As shown in sentiment graph 900, sentiment values ​​in nearby configurations are similar to each other. Figure 10 Examples show that multiple emotions such as "peace of mind", "stability", and "reassurance" have similar emotional values.

[0167] The above description primarily focuses on the functions of the data processing device 12 to illustrate the system disclosed herein. However, the system disclosed herein may not necessarily be implemented on a server. The system disclosed herein can be implemented as a general information processing system. For example, it can be implemented as a software program running on a personal computer, an application running on a smartphone, etc. The methods disclosed herein can be provided to users in the form of SaaS (Software as a Service).

[0168] In the above embodiments, an example of specific processing performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and distributed processing of specific processes by multiple computers, including computer 22, is also possible. For example, the data generation model 58 may also be provided in an external device of the data processing apparatus 12, where data is generated based on input data.

[0169] In the above embodiments, an example of storing a specific processing program 56 in memory 32 is illustrated, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 performs specific processing according to the specific processing program 56.

[0170] Alternatively, the specific processing program 56 can be stored in a storage device such as a server connected to the data processing device 12 via a network 54, and the specific processing program 56 can be downloaded and installed into the computer 22 upon request from the data processing device 12.

[0171] Furthermore, it is not necessary to store the entire specific process 56 in a storage device such as a server connected to the data processing device 12 via the network 54, nor is it necessary to store the entire specific process 56 in the memory 32; a portion of the specific process 56 may also be stored.

[0172] As a hardware resource for performing specific processing, various processors can be used. For example, a CPU can be listed as a processor; it is a general-purpose processor that functions as a hardware resource for performing specific processing by executing software, i.e., a program. Furthermore, as a processor, a dedicated circuit can be listed as a processor; it is a processor with a circuit structure specifically designed for performing specific processing, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit).

[0173] Every processor has built-in or connected memory, and every processor performs specific processes by using memory.

[0174] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware resources for performing a specific process can also be a single processor.

[0175] As an example of a single processor, the first type consists of a combination of one or more CPUs and software, which functions as a hardware resource to perform specific processing. The second type uses a processor, such as a System-on-a-chip (SoC), which implements the entire system functionality, including multiple hardware resources performing specific processing, using a single IC chip. In this case, the specific processing is implemented using one or more of the aforementioned processors that serve as hardware resources.

[0176] Furthermore, as the hardware architecture of these various processors, more specifically, circuits combining semiconductor elements and other circuit components can be used. Moreover, the specific process described above is merely an example. Therefore, it goes without saying that, without departing from the main point, unnecessary steps can be removed, new steps can be added, or the processing order can be changed.

[0177] The foregoing descriptions and illustrations are detailed explanations of the parts covered by this disclosure and are merely one example of this disclosure. For instance, the descriptions of the above-described structure, function, role, and effect are just one example of the structure, function, role, and effect of the parts covered by this disclosure. Therefore, it goes without saying that, without departing from the spirit of this disclosure, unnecessary parts can be deleted, new elements can be added, or replacements can be made to the foregoing descriptions and illustrations. Furthermore, to avoid confusion and facilitate understanding of the parts covered by this disclosure, explanations of technical common sense that does not require special explanation for implementing this disclosure have been omitted from the foregoing descriptions and illustrations.

[0178] All documents, patent applications and technical standards described in this specification are incorporated herein by reference as if they were specifically and individually described as incorporated by reference.

[0179] In addition, the following notes further explain the above explanation.

[0180] (Note 1) A system for supporting cross-border and cross-cultural virtual store operations is disclosed. The system includes a virtual store space generation department, a communication support department, and a purchase behavior analysis department. The virtual store space generation department receives attribute information from customers and employees worldwide, including language, cultural background, time zone, purchase history, products of interest, virtual avatar settings, and desired service style. Based on this information, it generates a virtual store space that allows cross-border and cross-cultural customers and employees to participate simultaneously. The communication support department uses generative AI to perform real-time multilingual translation of text or voice inquiries, product descriptions, and customer service responses between customers and employees, taking into account each user's cultural background, past communication history, national purchasing preferences, religious considerations, gender expression, and respect. The system automatically performs cultural adaptation processing, distinguishing between verbal and informal expressions and understanding the meaning of nonverbal expressions. Furthermore, it applies generative AI-based emotion recognition algorithms to analyze the emotional state of interactive messages and responses, visualizing customer satisfaction and purchase intentions. The purchase behavior analysis department accumulates customers' action history, chat history, product browsing history, purchase history, feedback history, questionnaire results, and rating scores in the virtual space as long-term memory. Through high-precision data analysis based on generative AI, it extracts popular products, analyzes purchase tendencies in different cultural circles, proposes optimization suggestions for service quality, automatically generates promotional plans, and provides prompts for employee training programs. Its key feature is that all these processes are comprehensively and autonomously executed through generative AI.

[0181] (Note 2) As described in Appendix 1, the system is characterized in that, in addition to customer and employee attribute information, the virtual store space generation unit also acquires external data such as holidays, events, seasons, local trends, laws and regulations, logistics status, exchange rate information, and popular product rankings in various regions. Based on this information, it dynamically optimizes the layout, product display, promotional display, reception style, event performances, and product descriptions within the virtual space, and uses generative AI to automatically provide virtual store experiences optimized for different countries and cultural spheres.

[0182] (Note 3) As described in Appendix 1, the system is characterized in that the communication support department uses a natural language processing function based on generative AI to perform real-time multilingual translation of messages between customers and employees. Furthermore, based on each user's cultural background, purchase history, past communication history, purchase tendency, and feedback content, it autonomously performs the following actions: selection of expression, switching between honorific and informal expressions, avoidance of cultural taboos, adjustment of the strength of emotional expression, automatic generation of expressions to enhance purchase intention, personalization of promotional messages, real-time visualization of customer satisfaction based on emotion recognition algorithms, and additional explanations or cross-selling and additional sales proposals performed by AI as needed.

Claims

1. A system for enabling global collaboration among multinational teams, comprising a virtual space generation department, a communication support department, and a project schedule management department. The virtual space generation department accepts access from multiple physically separated locations or user terminals, registers language and cultural background information as attribute information for each user, and generates virtual office spaces that the multinational team can participate in simultaneously based on this information. The communication support department uses generative AI to translate text or voice messages between users in real time, and performs cultural adaptation processing to automatically adjust expressions and subtle differences based on each user's cultural background. Furthermore, it applies a generative AI-based emotion recognition algorithm to analyze the emotional state of interactive messages, visualizing the quality and atmosphere of communication. The project schedule management department maintains the activity history of users within the virtual space and past project data as long-term memory, and through high-precision data analysis based on generative AI, it monitors progress, extracts tasks, optimizes task allocation, detects signs of cultural friction, and proposes mitigation strategies. The system is characterized by… All these processes are performed comprehensively and autonomously by generative AI.

2. The system as described in claim 1, characterized in that, The virtual space generation unit acquires network information and communication infrastructure information of existing physical office locations. Based on this information, it generates a metaverse space that virtually connects locations in various countries, allowing users to enjoy an experience as if they were in a real office, regardless of geographical limitations. Furthermore, using generative AI, it dynamically optimizes the layout and interaction within the virtual space based on the user's login status and activity status.

3. The system as described in claim 1, characterized in that, The communication support department uses natural language processing based on generative AI to perform real-time multilingual translation of messages between users. Furthermore, it automatically selects expressions, switches between honorific and informal expressions, avoids cultural taboos, and adjusts the strength of emotional expressions based on each user's cultural background and past communication history. In addition, based on the emotional state detected by the emotion recognition algorithm, it autonomously executes coordination or arbitration proposals made by AI as needed.

Images