Information processing system

Abstract

The application provides an information processing system. An information processing system comprises a processor configured to acquire and integrate open data and private data of an autonomous body; the processor is configured to calculate an evaluation benchmark using the integrated data; the processor is configured to automatically generate a report document according to a prompt using a generative artificial intelligence model.

Description

Technical Field

[0001] The technology disclosed herein relates to an information processing system. Background Technology

[0002] Japanese Patent Application Publication No. 2022-180282 discloses a method for controlling a role-based chatbot executed by at least one processor. The method includes the following steps: receiving a user's speech; adding the user's speech to a prompt word, the prompt word containing instruction statements associated with an explanation of the chatbot's role; encoding the prompt word; and inputting the encoded prompt word into a language model to generate a chatbot response to the user's speech.

[0003] Currently, reports required for scenarios such as policy initiation, subsidy applications, and project evaluations in autonomous regions largely rely on manual data collection, organization, and analysis from scattered data sources, followed by report writing according to a fixed format. This approach has the following problems: First, the open data of autonomous regions and the private data are stored in different systems and platforms, with heterogeneous data formats. Manual integration is labor-intensive, error-prone, and difficult to reflect the latest regional situation in a timely manner. Second, the evaluation benchmarks (such as KPIs or quantitative indicators used for projects such as payment, subsidies, and social capital preparation) are mostly set manually by experts, lacking systematic calculation and objective quantification of multi-source data, which makes it difficult to fully reflect regional characteristics and affects the accuracy of policy and fund allocation. Third, existing automated document generation tools are mostly based on simple template filling, which makes it difficult to combine large-scale integrated data for in-depth analysis and summarization, and cannot flexibly adapt to different format requirements and complex report structures, resulting in limited automation. Fourth, existing systems often ignore users' emotions and acceptance levels during use, and cannot dynamically adjust the expression and granularity of report content according to users' emotional state, making it difficult for reports to balance professionalism and readability, resulting in low efficiency in decision-making and communication.

[0004] Therefore, it is necessary to provide a system that can automatically acquire and integrate open data and civilian data from autonomous regions, automatically calculate evaluation benchmarks based on the integrated data, and automatically generate reports and supplementary materials in various formats by using generative artificial intelligence models while taking into account user sentiment, in order to improve the efficiency and accuracy of data analysis and report generation, enhance the ability of region-specific analysis, and improve user experience. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an information processing system. The system includes a processor configured to: acquire and integrate open data and private data from autonomous regions, wherein the open data includes statistical data on population, budget, infrastructure, and public services, and the private data includes consumer behavior data, business data, and transportation data, thereby forming a comprehensive dataset for a specific autonomous region under a unified data structure; calculate evaluation benchmarks using the integrated data, wherein the evaluation benchmarks include key performance indicators for subsidies, payments, social capital development projects, and other public policy projects; and automatically generate quantitative indicators reflecting the objective conditions and development goals of the target region through comprehensive analysis of historical data, multi-source data, and regional characteristics.

[0006] Furthermore, the processor is configured to combine data from a specific autonomous region with specific civilian data to achieve region-specific analysis. Specifically, based on autonomous region identification, project type, and policy needs, it automatically selects the datasets most relevant to the region and projects from multi-source data to perform customized analysis on the commercial vitality, traffic flow, population structure, etc. of a specific region, so that the calculated evaluation benchmark can reflect the differences and regional characteristics between autonomous regions.

[0007] Furthermore, the processor is configured to automatically generate reports and supplementary materials based on prompts using a generative artificial intelligence model. These prompts instruct the generative AI model to generate content within a specific report format, structure, and tone. The processor controls the chapter structure, table and chart insertion positions, and data citation methods of the reports and supplementary materials according to pre-set specific format or template requirements. Simultaneously, the processor adjusts the information presentation based on the user's emotional state, including adjusting the level of detail, word style, and positive or neutral expression tendencies of the explanatory text. For this purpose, it generates or updates prompts to instruct adjustments to the parameters or instructions of the generative AI model, thereby dynamically adjusting the output results.

[0008] Through the above structure and steps, the system of the present invention can automatically complete the acquisition and integration of multi-source data, the objective calculation of evaluation benchmarks, and the automatic generation of reports and supplementary materials in a predetermined format without a large amount of manual intervention. Furthermore, it can adjust the output content according to the user's mood, making policy analysis, payment application, and project evaluation at the autonomous region level more efficient, accurate, and targeted.

[0009] "System" refers to an overall device or platform consisting of hardware and / or software, including at least one processor and storage units, communication interfaces, etc. that work together with it, for performing functions such as data acquisition, data processing, and document generation.

[0010] A "processor" refers to a hardware or virtual computing unit that can execute program instructions and process and operate on data, including but not limited to a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), and virtual processing nodes running in a cloud environment.

[0011] "Autonomous entity" refers to a local administrative body that has autonomy under the legal framework of a specific country or region, including but not limited to local administrative units such as cities, districts, counties, and townships.

[0012] "Open data" refers to datasets that are provided publicly by public institutions such as autonomous regions, are machine-readable, and can be freely used, reused, and redistributed under certain conditions. These datasets include statistical data, geographic information, budgetary information, etc.

[0013] "Civilian data" refers to data generated or held by enterprises, groups or other non-governmental entities, including but not limited to consumer behavior data, business transaction data, transportation data, social media data, etc.

[0014] "Integration" refers to the process of collecting, cleaning, transforming, and unifying data from multiple sources to achieve consistency in structure, format, and semantics, thereby forming a comprehensive dataset that can be used for unified analysis and processing.

[0015] "Evaluation benchmark" refers to a set of quantitative indicators or indicators calculated based on integrated data to measure project performance, policy effectiveness, or the rationality of fund use, including key performance indicators (KPIs) and their benchmark values ​​and target values.

[0016] "Generative AI models" refer to AI models that can automatically generate text, images, or other content based on input prompts or conditions, including but not limited to large language models, text generation models, and multimodal generation models.

[0017] "Hints" refer to the input information provided to generative artificial intelligence models to instruct the model on what content to generate, as well as its style, structure, and focus, including natural language descriptions, control parameters, and format constraints.

[0018] "Report documents" refer to formal documents used to describe the project content, progress, use of funds, and evaluation results, including but not limited to application forms, implementation plans, interim reports, annual reports, and post-evaluation reports.

[0019] "Supplementary materials" refers to additional documents or materials provided to complement the report, including but not limited to appendices, charts and graphs, technical attachments, and data description documents.

[0020] "Data for a specific autonomous entity" refers to a set of data associated with a particular autonomous entity, including the autonomous entity's open data and, where necessary, historical project data, statistical data, etc., corresponding to that autonomous entity.

[0021] "Specific private data" refers to a collection of private data that is directly related to a specific autonomous region, region, or project, such as consumer behavior data, traffic flow data, or business transaction data collected within a specific autonomous region.

[0022] “Region-specific analysis” refers to personalized and differentiated data analysis based on a specific autonomous region and its corresponding multi-source data, targeting the region’s economic, population, transportation, and commercial characteristics, in order to reflect the characteristic differences between different regions.

[0023] "Specific format" refers to the document structure and layout required in advance or by relevant institutions, including chapter division, heading levels, table and chart insertion positions, writing style and terminology.

[0024] "User emotions" refers to the emotional state exhibited by the user or identified by the system during the interaction between the user and the system, including but not limited to positive, negative, neutral, anxious, conservative, and other emotional tendencies.

[0025] "Output results" refers to the overall content generated and presented to users by generative artificial intelligence models and system processing flows, including the main text of reports, supplementary information, charts, and related explanatory texts. Attached Figure Description

[0026] Figure 1 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the first embodiment.

[0027] 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.

[0028] Figure 3 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the second embodiment.

[0029] 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.

[0030] Figure 5 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the third embodiment.

[0031] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and head-mounted terminal according to the third embodiment.

[0032] Figure 7 This is a conceptual diagram illustrating an example of the configuration of the data processing system according to the fourth embodiment.

[0033] Figure 8 This is a conceptual diagram illustrating an example of the main functions of the data processing device and robot according to the fourth embodiment.

[0034] Figure 9 This represents an emotion map that maps multiple emotions.

[0035] Figure 10 This represents an emotion map that maps multiple emotions.

[0036] Figure 11 This is a sequence diagram illustrating the processing flow of the data processing system of the first embodiment.

[0037] Figure 12 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 1.

[0038] Figure 13 This is a sequence diagram illustrating the processing flow of the data processing system of the second embodiment.

[0039] Figure 14 This is a sequence diagram illustrating the processing flow of the data processing system in Application Example 2. Detailed Implementation

[0040] Hereinafter, an example of an implementation of the system according to the present disclosure will be described with reference to the accompanying drawings.

[0041] First, let me explain the terminology used in the following instructions.

[0042] In the following embodiments, the processor (hereinafter referred to as "processor") with reference numerals may be a single computing device or a combination of multiple computing devices. Furthermore, the processor may 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.

[0043] In the following embodiments, RAM (Random Access Memory), as indicated in the figures, is a memory that temporarily stores information and is used as working memory by the processor.

[0044] In the following embodiments, the memory, as indicated by the reference numerals, is one or more non-volatile storage devices that store various programs and parameters. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disks (e.g., hard disks), or magnetic tapes.

[0045] In the following embodiments, the communication I / F (Interface) with reference numerals is an interface that includes a communication processor and an antenna, etc. The communication I / F is responsible for communication between multiple computers. As an example of a communication specification applicable to the communication I / F, wireless communication specifications such as 5G (5th Generation Mobile Communication System), Wi-Fi (wireless fidelity) (registered trademark), or Bluetooth (registered trademark) can be listed.

[0046] In the following embodiments, "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 to express more than three items, the same interpretation as "A and / or B" applies.

[0047] First Implementation Method Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0048] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. A server can be cited as an example of the data processing device 12.

[0049] 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. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0050] The smart device 14 includes a computer 36, a receiving device 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. In addition, the receiving device 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0051] The receiving device 38 includes a touchscreen 38A and a microphone 38B, and receives user input. The touchscreen 38A receives user input via touch by detecting contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input via sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing the user input received by 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.

[0052] 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.

[0053] Communication I / F44 is connected to network 54. Communication I / F44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0054] Figure 2The diagram shows an example of the main functions of the data processing device 12 and the smart device 14.

[0055] 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 according to the specific processing program 56 executed on the RAM 30.

[0056] 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. In the emotion inference function (emotion-specific function) using the emotion-specific model 59, various inferences and predictions related to the user's emotions are performed, including inferences and predictions of the user's emotions, but this is not limited to this example. Furthermore, emotion inference and prediction may also include, for example, emotion analysis (parsing).

[0057] In the smart device 14, the processor 46 performs the acceptance output processing. The memory 50 stores the acceptance output program 60. The acceptance output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the acceptance output program 60 from the memory 50 and executes the read acceptance output program 60 on the RAM 48. The acceptance output processing is implemented by the processor 46 acting as the control unit 46A according to the acceptance output program 60 executed on the RAM 48. Furthermore, the smart device 14 has the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and these models can also be used to perform the same processing as the specific processing unit 290. The acceptance output processing is implemented by the processor 46 acting as the control unit 46A according to the acceptance output program 60 executed on the RAM 48.

[0058] Alternatively, other devices besides the data processing device 12 may also have the data generation model 58. For example, a server device (e.g., a generation server) may have the data generation model 58. In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with the server device that has the data generation model 58. Furthermore, the data processing device 12 may be a server device or a user-held terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing of the data processing system 10 of the first embodiment will be described.

[0059] Example 1 The flow of a specific process in Example 1 will be described. 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."

[0060] In the fields of public information processing and social capital construction planning, with the rapid increase in data volume from both public and private information sources, traditional solutions relying on manual processing and static reporting tools have revealed several shortcomings at the computer technology level in terms of computing resource utilization, data consistency maintenance, and support for multi-year planning. Specifically, firstly, existing systems are mostly based on a single data source or simple file-level integration, lacking a unified model management and type constraint mechanism for multi-source structured data. This makes it difficult to ensure the consistency of key indexes such as region and year at the database level, thus hindering the stable execution of large-scale automated analysis and subsequent intelligent generation processing in the computer. Secondly, existing systems that utilize artificial intelligence to generate text often employ static, manually written prompts, decoupling business data from generative AI models. They fail to deeply bind the structured indicator calculation results with the prompt generation logic at the system level, resulting in a lack of precise contextual constraints on model inputs. Consequently, the generated results are difficult to form predictable and reusable structured outputs within the computer, reducing the system's automation level and reliability. Third, existing report generation processes often separate statistical analysis modules from document generation modules, failing to design an end-to-end pipeline from the system architecture level, encompassing "analysis results → indicator setting → prompt statement generation → AI output → structured document assembly." This forces servers to undergo extensive manual intervention and multiple manual format conversions when processing multi-year financial planning, budget optimization, and execution evaluation, resulting in wasted computing resources and processing delays. Fourth, existing technologies generally lack a mechanism for dynamically constraining generative AI models using trend analysis and future forecast results. They cannot automatically inject time-series trend information and forecast intervals during the model invocation phase, causing the planning recommendations output by the model to be disconnected from the underlying database's computing capabilities in the time dimension, making it difficult to achieve machine-executable planning for medium- and long-term social capital construction.

[0061] Therefore, it is necessary to propose a technical solution to improve data management, indicator calculation, prompt statement generation, and automatic document assembly processes on the server side. From the perspective of computer architecture and data processing pipeline, this solution systematically improves the quality of multi-source data integration, the efficiency of automatic analysis, and the controllability of generative artificial intelligence model invocation. This enables the server to efficiently and stably generate multi-year financial planning and report documents with a higher degree of structure and strict correspondence with underlying indicators, thereby achieving the synergistic optimization and utilization of computing and data resources.

[0062] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Embodiment 1 is achieved by the following means.

[0063] In this invention, the server includes a device for acquiring structured data from public and private information sources, performing unified attribute processing and consistency constraint settings on the structured data using database management functions, and generating integrated data sets by region and year; a device for calculating evaluation indicators such as population growth rate, budget utilization efficiency, and their correlation based on the integrated data sets using tabular data processing software, setting social capital construction evaluation benchmarks or target values, and storing them in the form of indicator information; and a device for automatically or semi-automatically generating prompt statements for inputting a generative artificial intelligence model using the indicator information and summary or abstract information extracted from the integrated data sets, sending the prompt statements and background information to the generative artificial intelligence model to obtain descriptive information containing multi-year funding plans, budget allocation schemes, and performance evaluations, thereby automatically generating application documents, report documents, and accompanying tables or graphics in a predetermined format and outputting them electronically. This enables the formation of an integrated processing chain within the server, encompassing "unified modeling and constraint management of multi-source data—structured indicator calculation—generation of dynamic prompts for the model—assembly of structured documents based on AI output." At the computer level, this significantly reduces resource overhead caused by data inconsistency and manual intervention, improves the contextual accuracy and output controllability of generative artificial intelligence model calls, and achieves efficient automation of multi-year financial planning and report generation processes. Ultimately, this enhances the technical performance of the entire information processing system in terms of data integration capabilities, computational efficiency, and document generation quality.

[0064] A "system" refers to an overall technical solution consisting of at least one information processing device, a storage device, and a communication device, which executes data acquisition, storage, analysis, generation, and output processes through a program.

[0065] "Information processing device" refers to an electronic device with a processor and memory that can execute program instructions to acquire, process, convert, store and control the output of data, typically a server or computer terminal.

[0066] "Public information sources" refer to structured or semi-structured data sources provided by public institutions or public service providers, including information on population, finance, infrastructure, etc.

[0067] "Civilian information sources" refer to data sources provided by non-public entities such as enterprises, social organizations, or individuals, including information on consumer behavior, market trends, and industry activities.

[0068] "Structured data" refers to data organized with predefined fields, data types, and record formats, suitable for storage and processing in databases or spreadsheet software.

[0069] A relational database is a database system that organizes data using row and column structures and manages data based on a relational model. It is used to store and query structured data through tables and the relationships between them.

[0070] "Data management function" refers to software or modules used to perform operations such as data definition, insertion, update, deletion, query, and constraint control on storage media.

[0071] "Attribute unification" refers to the standardization of data field names, data types, encoding methods, and value ranges from different data sources, so as to enable unified management and calculation under the same data model.

[0072] "Consistency constraints" refer to the rules and conditions imposed on fields or records in a database to ensure data consistency, such as primary keys, foreign keys, uniqueness, and not null.

[0073] "Integrated dataset" refers to a combined dataset formed by integrating data from public and private information sources according to predetermined keys (such as region and year) under a unified model and constraints.

[0074] "Region" refers to a unit that can be used to identify geographical scope, such as national, regional, or local administrative divisions, serving as an index for data aggregation and analysis.

[0075] "Annual" refers to a time interval measured in calendar years or fiscal years, used to identify and distinguish statistical data from different periods.

[0076] "Table data processing software" refers to software tools or libraries that can load data in the form of tables or data frames and perform operations such as filtering, aggregation, statistics, calculation, association and pivoting.

[0077] "Evaluation indicators" refer to numerical parameters used to quantify the status or performance of an object, including population growth rate, budget utilization efficiency, and correlation with civilian data.

[0078] The "population growth rate" refers to the relative increase or decrease in the population of a certain region within a given time interval. It is generally obtained by dividing the difference between the current population and the previous population by the previous population.

[0079] "Budget utilization efficiency" refers to the ratio or function between the actual budget invested and the resulting public service effectiveness, population change or other effects within a given time period and region.

[0080] "Correlation index" refers to a statistical measure of the degree of correlation between two or more variables, such as a coefficient used to represent the degree of linear relationship between budget efficiency and consumption index.

[0081] "Social capital development" refers to the collective activities of planning, investing in, constructing, and operating public infrastructure and public service resources (such as transportation, education, and healthcare).

[0082] "Evaluation benchmark" refers to the reference value or range used to judge whether a certain evaluation indicator meets the standard or is good or bad, including threshold, target range or grade standard.

[0083] "Target value" refers to the expected value or range of values ​​set for a specific evaluation indicator in a certain time period or region, used to guide planning and evaluation.

[0084] "Indicator information" refers to a collection of information stored in the form of a data structure, including the name of the evaluation indicator, calculation results, target value, applicable region, and time range.

[0085] "Generative AI models" refer to AI models that are trained on large-scale data and can automatically generate text, tables, or other content based on input natural language or structured conditions.

[0086] "Prompt statements" refer to input text constructed to guide generative artificial intelligence models to produce specific types and formats of output. They typically contain background information, constraints, and output requirements.

[0087] "Background information" refers to the data summary, statistical results, or contextual description provided when a generative artificial intelligence model is invoked, which describes the current situation.

[0088] "Financial planning" refers to a plan that arranges and allocates the sources, uses, and amounts of funds for one or more future years.

[0089] A “budget allocation scheme” refers to a specific plan for allocating budget funds according to region, project, or field under a given total budget constraint.

[0090] "Performance evaluation" refers to the process and results of evaluating the effectiveness of implemented projects or budget usage based on pre-set evaluation indicators and target values.

[0091] "Application documents" refers to electronic or paper documents submitted to higher-level institutions or funding providers that contain planning content, funding requirements, and supporting explanations.

[0092] "Report document" refers to an electronic or paper document used to summarize, explain, or prove the work results, fund usage, and target achievement of a certain stage.

[0093] "Predefined format" refers to the pre-defined specifications for document structure, chapter order, table style, and graphic layout, which are used to unify the appearance and content organization of automatically generated documents.

[0094] A "table" is a structured representation of data displayed in rows and columns, used to clearly present multiple indicators and multi-dimensional data.

[0095] "Graphics" refers to images that visually represent the relationship between data and indicators, such as line charts, bar charts, pie charts, and scatter plots.

[0096] "Electronic documents" refer to document data stored in digital form on computer-readable media, including text, tables, images, and other content.

[0097] "Trend information" refers to descriptive information about the direction and speed of change of evaluation indicators over time, based on historical data.

[0098] "Future forecast interval" refers to the estimated range of possible values ​​for an indicator within a certain future time period based on a model or statistical method.

[0099] A "multi-year social capital development plan" refers to a comprehensive planning scheme that covers multiple years and arranges the investment priorities, budget scale and target indicators for each year.

[0100] "Risk factors" refer to uncertain events or conditions that may adversely affect the achievement of social capital construction goals, such as fiscal pressure and changes in population flow.

[0101] "Countermeasures" refers to a set of mitigation, avoidance, or response measures designed to address identified risk factors.

[0102] In one embodiment of the invention, the server includes a processor, main memory, non-volatile memory, and a network interface, while the terminal includes a processor, memory, a display device, and an input device (keyboard, pointing device, or touchscreen). The server and terminal communicate via a wired or wireless network. The server runs a server operating system, such as a Unix-like operating system, while the terminal runs a general-purpose operating system, such as a desktop or mobile operating system. The server runs a web application framework (e.g., a web framework based on a general-purpose scripting language), a task scheduler (e.g., a time-based task management tool or message queue framework), and a database management system (e.g., a relational database management system). The terminal runs a web browser (e.g., a general-purpose web browser) or a desktop client, as well as a data analysis environment (e.g., a distribution containing interactive notebooks and scientific computing libraries).

[0103] The server, at the program level, consists of multiple functional modules, including at least: a data receiving module, a data cleaning and pattern unification module, a data storage and constraint control module, an indicator calculation support module, a prompt statement generation module, a generative artificial intelligence invocation module, a document assembly and output module, and a trend analysis and prediction module. The terminal consists of a data download and upload module, a database access module, a visualization module, a prompt statement editing module, and a generative artificial intelligence result display module.

[0104] The server utilizes a database management system (such as a relational database management system) to create multiple logical tables in non-volatile memory. The server stores data such as population and budget from "public information sources" in the first table, consumer behavior and market trends from "private information sources" in the second table, user-defined evaluation benchmarks and target values ​​in the indicator table, and report generation records and generation resources (charts, etc.) in the report management table and resource table. The server defines primary keys, foreign keys, uniqueness constraints, and NOT NULL constraints on these tables. Primary keys typically consist of a region identifier (region_id), a year (year), and a data source type field. Foreign keys are used to establish referential relationships between the public data table, the private data table, and the indicator table. This relational schema design allows the database query optimizer to utilize indexes and constraints to reduce full table scans and shorten data access latency during large-scale queries and join operations, thereby improving data access efficiency.

[0105] The server's "Data Cleaning and Schema Unification Module" utilizes tabular data processing software libraries (such as data frame processing libraries) to perform field mapping and type conversion on uploaded structured data in main memory. The server unifies column names from different data sources to a standard field set; for example, it unifies different names like "Region Code" and "Administrative District Number" into "region_id," and unifies "Total Population" and "Resident Population" into "population." The server converts string fields representing years to integer types, fields representing amounts and indicators to floating-point types, and date strings to timestamps. The server also detects missing and outlier values, using rules (such as negative population or reversed year order) to mark abnormal records and store them separately in an error record table, rather than simply discarding them. This unified schema management and constraint control on the server side reduces the need for repeated cleaning on the terminal side, allowing subsequent indicator calculation modules to perform vectorized operations on a unified data structure. This reduces the number of branches in type checking and anomaly handling, thus lowering the CPU instruction branch error rate when processing large datasets, thereby improving overall processing speed and cache hit rate.

[0106] During the data analysis phase, the terminal accesses the server database using a data analysis environment and scientific computing libraries. The terminal reads data from the server's public and private data tables within a specified geographical and annual range via a database driver, and constructs data frames in local memory. The terminal performs sorting, grouping, and difference operations on the data to calculate indicators such as population growth rate and budget utilization efficiency. For example, using region as the grouping key and year as the time series key, the terminal calculates the population growth rate by dividing the population difference between consecutive years by the previous year's population. Based on a preset budget efficiency function, the terminal derives budget utilization efficiency by combining population growth rate, total budget, and private consumption activity indicators. By requesting consecutive years of data from the database at once and performing vectorized operations locally, the terminal significantly reduces the number of database round-trip communications and network load compared to requesting and calculating data line by line, thereby increasing the overall throughput during the indicator calculation phase.

[0107] In some implementations, the server can also directly perform the above indicator calculations on the server side. In this case, the server periodically triggers batch processing jobs through its internal task scheduling module, calling the data frame processing library to perform batch calculations on the aggregated data set in the database. The server loads the time series data of each region into a memory data frame, uses grouping and aggregation methods to calculate statistics such as the average population growth rate, minimum budget utilization efficiency, and standard deviation over the past N years, and writes these indicator results back to the indicator table. The server can also calculate correlation indicators, such as using the Pearson correlation coefficient or Spearman's rank correlation coefficient to calculate the correlation between budget utilization efficiency and the private consumption index. By centrally executing these calculations on the server, the system can leverage the server's multi-core CPU and large memory advantages to process data from different regions in parallel partitioning, thereby completing data analysis for more regions and years within the same time window, resulting in technical improvements in processing throughput and response time.

[0108] The server's "Trend Analysis and Prediction Module" utilizes time series analysis algorithms or deep learning-based time series prediction models to estimate the trends and future intervals of evaluation indicators. In one implementation, the server employs an autoregressive integrated moving average model or an exponential smoothing model to fit the time series of each indicator and calculate predicted values ​​and confidence intervals for several future years. In another implementation, the server uses a recurrent neural network or a long short-term memory network as the time series prediction model. The server inputs historical indicator vectors into the network in chronological order. The network consists of multiple hidden layers, each with several neurons, using a non-linear activation function, and outputs predicted indicator values ​​for several future time steps. During the training phase, the server minimizes the mean squared error loss function on historical data, updating the network weights using gradient descent or its variants (such as adaptive learning rate optimization algorithms). During the prediction phase, the server uses the trained weights to slide the input historical sequence in a window manner, thereby generating future prediction intervals. These trend and prediction results are written by the server into an indicator table and an auxiliary table, used as part of the constraints and background information when generating subsequent prompts.

[0109] The server's "prompt statement generation module" in this invention features structured construction capabilities, rather than relying on manually written free text. The server automatically extracts the following information from the indicator table and aggregated data set: which regions have population growth rates lower than the evaluation benchmark, which regions have budget utilization efficiency significantly lower than the overall average, and which sectors have a high correlation between investment and private consumption activity. Based on the configured rule template, the server fills this information into the slots of the prompt statement template, generating structured natural language input for generative artificial intelligence models. An example of a prompt statement used by the server is as follows: "Based on the following calculated social capital preparation indicators:" - Region A, average population growth rate from 2020 to 2024: 1.2%, budget utilization efficiency: 0.65; - Region B, average population growth rate from 2020 to 2024: 1.8%, budget utilization efficiency: 0.82; - Region C, average population growth rate from 2020 to 2024: 0.9%, budget utilization efficiency: 0.58; Please use a generative artificial intelligence model to generate an optimized allocation of the 2025 budget.

[0110] Require: 1. Provide suggested budget increase / decrease percentages for each region; 2. Explain the reasons for the adjustments, with a focus on improving population growth rate and the quality of infrastructure services; 3. Output the results in list or table format. When generating such prompts, the server organizes information according to a predefined non-natural language hierarchical structure (such as an internal object tree or key-value pair structure) and then linearizes it into a natural language description. Therefore, the prompts are not manually written all at once, but rather automatically assembled by the server according to rules. These rules may include: sorting regions by the degree of deviation from the index, selecting only the most severe regions, and limiting the output format to "tables" or "bars," etc. Through this structured prompt generation process, the server can precisely control the context capacity and key areas input to the generative AI model, thereby reducing unnecessary information redundancy and the search space during model decoding. This helps improve the relevance and consistency of the output content, which technically manifests as increased effective token utilization in model calls, reduced generation time, and lower error rates.

[0111] In one implementation, the generative AI model invoked by the server is a language model based on a transformer architecture. The server stores model parameters in the model deployment environment. This model includes multi-layered self-attention encoders and decoders, each layer comprising a multi-head self-attention sublayer and a feedforward network sublayer. During inference, the server encodes prompts as vector sequences, which are then processed by an embedding layer and positional encoding before being input into the model. The model uses a self-attention mechanism to calculate attention weights between tokens at different positions, generating context-sensitive representations, which are then output as the probability distribution of the next token via a feedforward layer. The server employs a greedy decoding or bundle search decoding strategy, selecting the token sequence with the highest probability from the output probability distribution as the generated result. During the model's training phase (completed in a separate computational environment), the server uses cross-entropy loss as the error function, updates network weights using stochastic gradient descent and backpropagation, and employs data augmentation methods during training, such as paraphrasing, perturbation, and noise injection of the input text, to enhance the model's robustness to different representations. This specific model structure and training method enable the generated results to have good performance in semantic consistency and structured expression, thereby allowing the subsequent document assembly module to parse and embed them with fewer rules, reducing the reliance on manual post-processing.

[0112] When the terminal interacts with the generative artificial intelligence model, users can edit and adjust the pre-generated prompts on the server through the interface to add business context or strategy preferences. For example, a user can enter the following prompt on the terminal: "Assuming the policy objective is:" - Average annual population growth rate in all regions ≥ 1.5%; - Infrastructure budget utilization efficiency in all regions ≥ 0.80; Based on existing data and indicators, please develop a social capital investment strategy for the next three years using a generative artificial intelligence model, including: 1. Recommended budget range for annual investment; 2. Priority investment areas (such as transportation, education, healthcare, digital infrastructure, etc.) and the reasons; 3. Potential risks and corresponding countermeasures (e.g., fiscal pressure, population outflow, etc.). The terminal sends the user-edited prompts along with a summary of metrics provided by the server to the generative AI model interface. Upon receiving the model output, the terminal performs structured parsing of the table paragraphs, column paragraphs, and explanatory paragraphs. For example, the terminal can split "Region A Recommendation," "Region B Recommendation," etc., based on specific delimiters or heading patterns, and extract the budget change ratio and expected results into key-value pairs for direct reference by the server's document assembly module. This parsing process is automated through fixed rules, avoiding manual copying and pasting of text by the user and reducing human error and formatting inconsistencies.

[0113] The server's "Document Assembly and Output Module" employs a template-driven approach when generating application and report documents. The server stores document templates in its memory, with placeholders marking chapter titles, natural language paragraphs, table positions, and graph positions. The server sequentially fills the template with statistical tables retrieved from the database, suggestion tables parsed from the generative AI model, and pre-generated graphs, according to the placeholders. The server uses a document generation engine to render the templates and data into a formatted document (e.g., PDF or structured markup document), saving the output file to the file system or pushing it to a terminal for download. Because the table and graph data come from a unified database and analysis results, while the text suggestions come from the generative AI model, but are all inserted through a unified template engine, this integrated assembly avoids the field omissions and numerical errors caused by manual copying and pasting, improving the consistency and traceability of the generated documents. From a technical perspective, the server can automatically generate multiple structurally consistent and content-corresponding documents for different years and regions in a short time, significantly reducing I / O and processing time.

[0114] The system of this invention is not merely a simple replacement for manual paperwork, but a holistic optimization of the computer's internal data management structure, analysis algorithm flow, and generative artificial intelligence invocation method. By introducing a unified data model and consistency constraints on the server side, the system reduces the overhead of anomaly handling caused by non-standard data. By using vectorized data frame operations to replace loop-based script calculations on the server or terminal side, the system achieves superior performance compared to traditional script processing in terms of CPU cache utilization and instruction pipeline. Through structured prompt generation and model input constraints, the system enables the decoding process of generative artificial intelligence to run within a smaller search space, improving generation speed and reducing the proportion of irrelevant content. Through trend analysis and prediction modules, the system incorporates machine-calculated future interval information into the model input, tightly coupling the planning suggestions generated by the model with the underlying time series operations, fundamentally enhancing the rationality and verifiability of the model output in the time dimension.

[0115] In other implementations, the server can be replaced by a multi-node cluster, the database can be replaced by a distributed columnar storage system, the metric calculation can be performed in parallel using a distributed data processing framework, and the generative artificial intelligence model can be deployed on dedicated acceleration hardware (such as a graphics processing unit or tensor processing unit) to further improve inference speed. The prompt generation rules can also be extended to automatically derive constraints using a rule engine or logic program, thus eliminating the need for manual template modification when adding new metrics or regions. The terminal can be a mobile terminal; in scenarios with limited network bandwidth, the terminal only requests pre-aggregated results, reducing the amount of original data transmission and thus lowering the communication load. These variations, without altering the basic concept of the invention, further enhance the system's adaptability and technical effectiveness under different hardware environments and load conditions.

[0116] use Figure 11 The processing procedure is explained.

[0117] Step 1: Users obtain raw data files from public and private information sources using their terminals.

[0118] Input: Online resources such as population data, budget data, consumer behavior data, and market trend data from public and private information source websites.

[0119] Output: Structured data files (CSV or Excel files) saved in the terminal's local storage.

[0120] Users access public institution portals and private data service platforms via a web browser on their terminals, select a specific region, year, and data type, and click the download button. The terminal calls the operating system's network stack to receive the file stream via HTTP / HTTPS protocol and calls the file system interface to write the received data to the local disk, for example, generating "public_data.csv" and "private_data.xlsx". After the writing is complete, the terminal uses file reading functions to attempt to open the file and check the number of rows, fields, and encoding format. If file corruption or missing fields are detected, the user is prompted to download again on the display device.

[0121] Step 2: The terminal uploads the raw data file to the server.

[0122] Input: The CSV or Excel data file saved locally in the terminal in step 1.

[0123] Output: The raw data file stored in the server's temporary directory.

[0124] On the terminal interface, the user selects the desired file on the "Data Upload" page and clicks the upload button. The terminal constructs a multi-part form request in its event response, writing the selected file content as binary data into the HTTP request body, and sends it to the server's upload interface URL via the network interface. Upon receiving the request, the server caches the file in a memory buffer, then writes the file to a temporary directory in the server's non-volatile storage via the operating system's file I / O interface, and returns a successful upload response. The terminal parses the server's response status code and displays the upload progress and result on the interface.

[0125] Step 3: The server performs format parsing, field mapping, and data type conversion on uploaded files.

[0126] Input: The original CSV or Excel file in the server's temporary directory from step 2.

[0127] Output: A standardized data table structure (memory data frame) generated in main memory.

[0128] The server uses a spreadsheet data processing software library to read the file, parsing each row into a record and each column into a field. Based on a pre-defined field mapping table, the server renames columns from different sources to standard field names (e.g., unifying "Region Code" and "Administrative Area Code" as "region_id", and "Total Population" and "Resident Population" as "population"). The server performs data type conversions on each field: converting strings representing years to integers, converting monetary and index fields to floating-point numbers, and converting dates to timestamps. During this process, the server sets unparseable cells as missing values ​​and logs parsing error information. Through this unified format and type conversion, the server transforms heterogeneous formats from different data sources into a standardized in-memory data structure that can be uniformly processed in subsequent steps.

[0129] Step 4: The server performs data cleaning, anomaly detection, and consistency checks.

[0130] Input: The standardized memory data frame generated in step 3.

[0131] Output: The cleaned dataset that meets the constraints and the anomalous dataset recorded separately.

[0132] The server checks for missing values ​​in key fields (region_id, year, population, budget_total, etc.), marking records with missing key fields as anomalies and moving them to an anomaly set. The server applies rules to detect outliers, such as whether the population is negative, the budget amount is negative, or the year is reversed or repeated; records that do not meet these rules are also moved to the anomaly set. The server also iterates through records to check for duplicates of the same combination of region_id and year; if duplicates are found, it chooses to retain the latest record or merge them using the average, based on configured rules. Through these data processing operations, the server ensures that as much valid data as possible is retained while eliminating erroneous data, thereby reducing error propagation in subsequent calculations and improving the accuracy of indicator calculations.

[0133] Step 5: The server writes the cleaned data into a relational database and creates indexes and constraints.

[0134] Input: The cleaned dataset obtained in step 4.

[0135] Output: Common and private data tables stored in a relational database, along with related index and constraint information.

[0136] The server connects to the relational database management system and checks if the target data table already exists. If not, it creates a public and private data table based on a predefined schema, including field definitions, primary key, and foreign key design. The server inserts the cleaned data into the corresponding tables in batches. Before insertion, it checks for records with the same `region_id` and `year` based on the primary key to determine whether to perform an insert or update operation. The server creates a composite index for the `region_id` and `year` combination to optimize the performance of subsequent region- and year-based queries; it also sets reference constraints for foreign key fields to ensure consistency across table joins. This step, through structured storage and index design, enables efficient retrieval and updating of large-scale data.

[0137] Step 6: The terminal reads the data required for integration and analysis from the server database.

[0138] Input: The geographical range, year range, and indicator type specified by the user on the terminal; public data tables and private data tables in the server database.

[0139] Output: A merged data frame in the terminal memory (containing fields such as population, budget, and consumption index).

[0140] Users select the target region and time range on the terminal interface. The terminal then constructs an SQL query statement and sends the query request to the server via the database driver. The server executes the query in the database, quickly locates records that meet the conditions using indexes, and returns a result set. After receiving the result set, the terminal constructs a data frame locally and uses a merge operation to associate public and private data into a single table structure by region_id and year, preparing input for subsequent local calculations.

[0141] Step 7: Terminal computing uses basic evaluation indicators such as population growth rate, budget utilization efficiency, and relevance.

[0142] Input: The merged data frame in the terminal memory in step 6.

[0143] Output: An expanded data frame with added population growth rate field, budget utilization efficiency field, and correlation statistics.

[0144] The terminal first sorts the data by region_id and year, then groups it by region_id, and calculates the difference and ratio between previous and subsequent years for the population field to obtain the population growth rate for each region in each year. Based on a predefined budget efficiency function, the terminal combines population_growth_rate with budget_total and private consumption indicators to calculate budget utilization efficiency, for example, generating a dimensionless efficiency value in the form of "population growth rate / (total budget / constant)". The terminal can also perform correlation calculations between budget utilization efficiency and the consumption index to obtain the overall correlation coefficient and significance level. Through these vectorized operations, the terminal forms multidimensional evaluation indicators within a data structure, serving as the basis for subsequent target value setting and trend analysis.

[0145] Step 8: Users set evaluation benchmarks and target values ​​for social capital construction on the terminal.

[0146] Input: Various evaluation metrics in the extended data frame in step 7, as well as the charts and statistical summaries displayed on the terminal.

[0147] Output: Structured indicator setting information (indicator name, target value, applicable region and time range, etc.).

[0148] Users can view historical trends in population growth rates and budget utilization efficiency for various regions on the terminal, comparing them with the national average. Users set target values ​​for indicators based on policy intentions, such as setting a target population growth rate of 1.5% and a target budget utilization efficiency of 0.8. The terminal converts these user settings into structured objects, including indicator names, target values, assessment periods, and applicable regional scope. The terminal then prepares these indicator settings as data input to send to the server.

[0149] Step 9: The terminal submits the indicator settings to the server, and the server stores the indicator information.

[0150] Input: The structured index setting information generated in step 8.

[0151] Output: Records stored in the metrics table of the server database.

[0152] The terminal sends the metric settings in a structured format to the server via an HTTP interface. Upon receiving the request, the server parses out the name, target value, and description of each metric and writes them to the metric table in the database. The server checks if a metric with the same name already exists, updates or inserts a new record as needed, and creates an index for the `indicator_name` field in the metric table to speed up name-based queries. In this step, the server transforms the user's strategic goals into persistent data that can be referenced in subsequent calculations and generation.

[0153] Step 10: The server performs trend analysis and future predictions based on the integrated data set and indicator information.

[0154] Input: Public data tables, private data tables, and indicator tables in the database.

[0155] Output: A set of trend forecasts for each indicator across multiple future years, including predicted values ​​and confidence intervals.

[0156] The server invokes the trend analysis module to read time-series data of indicators for each region and year from the database, including historical population growth rates and budget utilization efficiency. The server fits a time-series model or deep learning prediction model to each time series, calculating predicted values ​​for several future years. Simultaneously, the server calculates the prediction error or confidence interval and writes the results, along with historical data, into the trend results table. The output of this step is the "trend information" and "future prediction interval" referenced in subsequent prompts, which can be used to constrain the planning content generated by the generative artificial intelligence model, ensuring consistency with the machine-calculated predictions.

[0157] Step 11: The server automatically generates structured prompts.

[0158] Input: A collection of aggregated data, indicator tables, and trend prediction results from the database.

[0159] Output: Prompt text for generative artificial intelligence models.

[0160] The server filters regions from the aggregated dataset that have lower-than-target population growth rates and budget utilization efficiency, and sorts them by the degree of deviation. For each region, the server extracts the average values ​​for recent years, the latest annual values, and trend forecasts. Based on a predefined language template, the server fills these values ​​into a natural language description, generating a prompt. For example, the server generates the following: "Based on the following calculated social capital preparation indicators:" - Region A, average population growth rate from 2020 to 2024: 1.2%, budget utilization efficiency: 0.65; - Region B, average population growth rate from 2020 to 2024: 1.8%, budget utilization efficiency: 0.82; - Region C, average population growth rate from 2020 to 2024: 0.9%, budget utilization efficiency: 0.58; Please use a generative artificial intelligence model to generate an optimized allocation of the 2025 budget.

[0161] Require: 1. Provide suggested budget increase / decrease percentages for each region; 2. Explain the reasons for the adjustments, with a focus on improving population growth rate and the quality of infrastructure services; 3. Output the results in list or table format. During the generation process, the server controls the length and filters the content of the prompt statements to ensure that the most important indicator information and constraints are included within a limited length.

[0162] Step 12: The terminal can edit or supplement the prompt statement as needed and submit the call request.

[0163] Input: The initial prompt statement generated by the server in step 11 and the policy descriptions or preferences added by the user; generative artificial intelligence service interface information.

[0164] Output: A final prompt request sent to the generative artificial intelligence model.

[0165] The terminal displays server-generated prompts on the interface, allowing users to add global policy objectives or specific constraints, for example: "Assuming the policy objective is:" - Average annual population growth rate in all regions ≥ 1.5%; - Infrastructure budget utilization efficiency in all regions ≥ 0.80; Based on existing data and indicators, please develop a social capital investment strategy for the next three years using a generative artificial intelligence model, including: 1. Recommended budget range for annual investment; 2. Priority investment areas (such as transportation, education, healthcare, digital infrastructure, etc.) and the reasons; 3. Potential risks and corresponding countermeasures (e.g., fiscal pressure, population outflow, etc.). The terminal packages the edited text together with a summary of background data provided by the server and sends it as a request to the API endpoint of the generative artificial intelligence model via the network interface.

[0166] Step 13: The server or terminal invokes the generative artificial intelligence model and obtains planning suggestions.

[0167] Input: The final prompt text generated in step 12 and the server-side API parameters (model name, decoding strategy, etc.).

[0168] Output: Text results including budget allocation recommendations, investment priorities, and risk analysis.

[0169] The server or terminal encodes the prompt statement into request data and sends it via HTTPS to an inference server deployed with a generative artificial intelligence model. Based on the structured information and constraints in the prompt statement, the model generates text containing multiple parts, including regional budget increase / decrease ratios, expected population growth, priority investment areas, and risk warnings. Upon receiving the returned results, the server or terminal breaks down the entire text into paragraphs and formatting marks, identifies tables, columns, and explanatory paragraphs, and extracts the structured information into an internal data structure for subsequent document assembly and chart generation.

[0170] Step 14: The server outputs assembly application documents and report documents based on database statistical results and generative artificial intelligence models.

[0171] Input: The integrated data set in the database, the indicator table, the trend forecast results, and the structured data of the planning suggestions parsed in step 13.

[0172] Output: Electronic document files conforming to a predefined format (e.g., PDF or markup document).

[0173] The server reads the document template and identifies data table placeholders, chart placeholders, and natural language paragraph placeholders within it. The server fills the table placeholders with statistical tables calculated from the database, fills the chart placeholders with pre-generated population growth rate trend charts and budget efficiency bar charts, and embeds recommendations parsed from the generative artificial intelligence model output into the corresponding sections. The server generates a complete document containing multi-year funding applications, performance reviews, and risk countermeasures, saves the document in the file system, and updates the report management table to record the document path and generation time.

[0174] Step 15: The terminal acquires and displays the final document and auxiliary visualization results.

[0175] Input: The electronic document file generated by the server in step 14 and its access path.

[0176] Output: Document content and visualizations displayed on a terminal display device, or a locally stored copy of the document.

[0177] Users request a list of the latest documents through the terminal's "Report Browsing" interface. The terminal retrieves the file information from the server and displays it. After selecting a target document, the terminal downloads the document file via the network and opens it locally. Users can then view automatically generated tables, charts, and text descriptions. The terminal can also display some charts interactively, allowing users to further examine detailed indicators for different regions and years. This output not only serves as a reference for decision-makers but also as a basis for subsequent model fine-tuning and rule adjustments, thus forming a closed-loop technical process between data, models, and documents.

[0178] Application Example 1 The process flow corresponding to the specific processing in 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".

[0179] In the technical field of utilizing publicly available information from public institutions and information held by private entities for regional analysis and policy formulation, existing technologies generally suffer from the following problems. First, when processing multimodal data (such as demographic data, behavioral data, event data, and financial data) from different sources and formats, servers often only perform simple summarization or static statistics, lacking mechanisms for fine-grained alignment and deep fusion based on temporal and spatial dimensions. This results in inconsistent and incomplete data foundations for subsequent analysis, making it difficult to support high-precision predictions and decision-making. Second, when building predictive models and calculating policy indicators, servers often employ fixed rules or single algorithms, limiting their ability to estimate the probability of events occurring in spatiotemporal regions. They cannot adaptively update models and indicator systems based on feedback data, thus failing to reflect changes in the real environment in a timely manner. Third, when generating reports, applications, and supplementary information, servers typically rely on manual editing or simple template filling, resulting in low automation. Furthermore, the generation process is disconnected from front-end data processing and model prediction results, making it impossible to use structured data to finely control generative artificial intelligence models. This leads to a mismatch between output content and management needs and user attributes, impacting information utilization efficiency. Finally, server notifications to external terminal devices and feedback of implementation results from terminals are often viewed as one-way information transmission. There is a lack of a systematic design to loop implementation feedback back into data processing and model updates, thus hindering continuous iteration to improve prediction performance and the quality of prompts. This limits the overall computing system's intelligence and scalability in resource allocation and countermeasure formulation. Therefore, it is necessary to provide an end-to-end computing mechanism encompassing data acquisition, fusion, modeling, indicator calculation, prompt generation, document generation, and closed-loop updates. This mechanism would improve the server's automatic analysis and intelligent generation capabilities in complex, multi-source data environments, thereby enhancing the performance and functionality of computer technology itself.

[0180] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Application Example 1 is achieved by the following means.

[0181] In this invention, the server includes a device for acquiring publicly available information from public institutions and information held by private entities via communication, aligning the information based on time and location information, and storing the aligned information as fused data; a device for extracting multiple attribute information from the fused data and performing statistical and preprocessing to generate a learning dataset and a reasoning dataset; a device for constructing a machine learning model based on the learning dataset and calculating the probability of events occurring in a spatiotemporal region based on the reasoning dataset, thereby calculating countermeasure indicators and evaluation benchmarks; a device for generating the countermeasure indicators and evaluation benchmarks into machine-readable structured data and generating prompt statements containing the structured data; and a device for inputting the prompt statements and the structured data into a generative artificial intelligence model to automatically generate document information, and updating the fused data, the machine learning model, the prompt statements, and the evaluation benchmarks based on the implementation results information from an external terminal device. This allows for the formation of a closed-loop computational chain within the server, encompassing data acquisition, spatiotemporal alignment, model training and online inference, strategy indicator calculation, structured prompt statement construction, controlled invocation of generative artificial intelligence models, and adaptive updates based on implementation feedback. This improves the efficiency of multi-source heterogeneous data processing at the computer implementation level, enhances the accuracy and stability of spatiotemporal event prediction, enables fine-grained control and personalized optimization of the output content of generative artificial intelligence models, and ultimately improves the overall automation level, resource allocation rationality, and operational performance of the information processing system.

[0182] "Information disclosed by public institutions" refers to data and information released to the public by various public management entities through online platforms, data interfaces, or documents, including but not limited to population statistics, financial information, event statistics, and infrastructure information, which are generally accessible and not targeted at any specific user.

[0183] "Information held by non-public entities" refers to data and information generated and stored by various organizations or individuals that are not public management entities during their business activities or equipment operation, including but not limited to behavioral data, monitoring device information, transaction information, service usage records, etc., which are usually subject to access restrictions or specific purposes.

[0184] "Time information" refers to the identification information used to indicate the time when data was generated or recorded, including but not limited to date, time, timestamp, time interval, etc., which is used to align, sort or aggregate data from different sources in the time dimension.

[0185] "Location information" refers to the identification information used to represent the spatial location of the data, including but not limited to address information, geographic coordinates, area codes, spatial grid numbers, etc., which are used to correspond and associate data from different sources in a spatial dimension.

[0186] "Alignment processing" refers to the process by which a server matches, normalizes, and reorganizes multiple sets of data from different sources and in different formats based on association keys such as time information and location information, so that data from the same time period and the same spatial region can be associated with each other in a unified data structure.

[0187] "Integrated data" refers to a comprehensive data set that, after alignment processing, is unified across multiple sources in both time and space dimensions and stored in the same data structure. This data set contains multiple attribute fields and serves as the basis for subsequent analysis and modeling.

[0188] "Attribute information" refers to various data items extracted from fused data to characterize the features of objects or events, including but not limited to different categories of feature data such as population information, behavioral information, event occurrence information, and financial information.

[0189] "Population information" refers to statistical data related to the size of a specific region or population, including but not limited to population size, age structure, gender ratio, and residential density, which are used to reflect the population characteristics of a region.

[0190] "Behavioral information" refers to data related to the activity status of an individual or group within a specific time and space, including but not limited to pedestrian flow, dwell time, number of times passing through entrances and exits, and movement trajectory characteristics, which are recorded by sensors, monitoring devices, or business systems.

[0191] "Event occurrence information" refers to data recording specific categories of events that occur at a specific time and location, including but not limited to the time, location, and type of occurrence of violations, accidents, and public security incidents.

[0192] "Funding information" refers to data related to fund allocation, fund application, fund use and its effects, including but not limited to budget figures, execution amounts, subsidy allocations, project expenditure structure, etc.

[0193] "Statistical processing" refers to various mathematical operations and summarization operations performed on attribute information to analyze data distribution and relationships, including but not limited to counting, summation, average, maximum, minimum, variance calculation, and grouping aggregation.

[0194] "Preprocessing" refers to various data processing operations performed before attribute information is input into machine learning models or other analysis modules to improve data quality and algorithm performance, including but not limited to missing value handling, outlier handling, standardization, normalization, encoding conversion, and feature construction.

[0195] "Learning dataset" refers to a set of sample data used to train machine learning models after statistical processing and preprocessing. This set includes input feature data and corresponding target labels or target values.

[0196] "Inference dataset" refers to a dataset constructed based on the same or similar feature structure as the learning dataset, used for prediction or inference by a trained model. This dataset typically does not contain or use target labels.

[0197] A "machine learning model" is a mathematical or statistical model that is automatically built by a learning algorithm based on a set of learning data. It is used to represent the mapping relationship between input features and output targets, and can predict or classify unknown samples.

[0198] "Spatiotemporal region" refers to a data analysis unit that is defined by both time and space, including but not limited to a combination of a specific time period and a specific geographical region, used to estimate and display the probability of an event occurring in the spatiotemporal dimension.

[0199] "Event probability" refers to the numerical value output by a machine learning model for a specific spatiotemporal region, representing the likelihood of a specific type of event occurring within a certain period of time in that spatiotemporal region. It is usually expressed in the form of probability or risk score.

[0200] "Countermeasure indicators" refer to a set of quantitative indicators calculated based on the probability of an event and related attribute information, used to guide resource allocation and the implementation of measures, including but not limited to patrol behavior indicators, monitoring device configuration indicators, and fund allocation indicators.

[0201] "Patrol behavior indicators" refer to quantitative parameters used to represent the amount and pattern of patrol activities that should be carried out in a specific time and space area, including but not limited to patrol frequency, patrol time period, number of patrol routes, personnel configuration, etc.

[0202] "Monitoring device configuration index" refers to quantitative parameters used to indicate the quantity and layout of monitoring devices that should be configured or adjusted in a specific time and space area, including but not limited to the number of monitoring devices, installation location density, coverage area, etc.

[0203] "Fund allocation indicators" refer to quantitative parameters used to represent the allocation of financial resources among multiple regions or projects, including but not limited to the recommended funding amount, funding priority, and investment ratio for each region.

[0204] "Evaluation benchmark" refers to a set of quantitative or qualitative standards used to measure the overall effectiveness of policy implementation and the rationality of resource allocation, including but not limited to policy indicator values, evaluation indicator values, and fund utilization indicator values.

[0205] "Evaluation index value" refers to a specific numerical value used as part of the evaluation benchmark to measure the system performance or the effectiveness of countermeasures, such as the change in event occurrence rate, risk level distribution, and resource utilization efficiency.

[0206] "Capital utilization index value" refers to a quantitative value used to evaluate the relationship between capital investment and expected results, including but not limited to the reduction in unit capital events, capital execution rate, and benefit-cost ratio.

[0207] "Structured data" refers to data organized according to a predetermined data pattern or data structure, with clearly defined fields and types, including but not limited to tabular data, tokenized data, key-value pair data, etc., which are suitable for automatic parsing and processing by computer programs.

[0208] "Prompt statements" refer to text content used to provide task descriptions and contextual information to generative artificial intelligence models, including but not limited to instructions on output format, content scope, tone and style, key items, and descriptions of embedded structured data.

[0209] "Generative AI models" refer to AI models built on machine learning methods that can automatically generate natural language text or other content based on input prompts and related data. They typically employ deep learning structures and possess language modeling and content generation capabilities.

[0210] "Natural language text" refers to continuous text content written in human natural language and output by generative artificial intelligence models, including but not limited to explanatory texts, reports, applications, and abstracts.

[0211] "Document information" refers to a collection of text and its accompanying content organized according to a predetermined format or standard, including but not limited to report information, application information, supplementary information, and corresponding titles, paragraph structures, lists, charts, and descriptions.

[0212] "Report information" refers to the document content used to explain the results of data analysis, event predictions, countermeasures, and their expected effects, and is usually used for internal management or decision support.

[0213] "Application information" refers to the document content used to submit applications for funding, projects, or resources to higher-level organizations or relevant entities, including the reasons for the application, the supporting data, the plan content, and the expected results.

[0214] "Supplementary information" refers to auxiliary documents used to supplement report and application information, including explanatory details, notes, background materials, appendix data, etc.

[0215] "Notification information" refers to message content generated by the server and sent to external terminal devices to alert users of event risks, provide countermeasure suggestions, or offer operational guidance. This includes, but is not limited to, text notifications, alarm messages, and task instructions.

[0216] "External terminal device" refers to a computing terminal that communicates with a server via a network and is used to receive, display, or provide feedback on notification information, including but not limited to mobile terminals, fixed terminals, and control devices.

[0217] "Implementation Result Information" refers to the data and information returned to the server by the external terminal device after the actual implementation of the countermeasures, including but not limited to the implementation status of the countermeasures, the resource input status, and the actual occurrence of events, which are used to evaluate and update the model and indicators.

[0218] In one embodiment of the invention, the server is deployed in a data center or cloud computing environment. The server hardware includes a central processing unit, a semiconductor storage device, and a network interface device. The server may employ a general-purpose processor with multi-core processing capabilities and may optionally be equipped with a graphics processing unit to accelerate deep learning inference and visual feature extraction. The basic software environment on which the server runs includes an operating system (e.g., a UNIX-like operating system), a relational database management system (e.g., general-purpose database software), a script execution environment, and an application server. The server interacts with multiple terminals via a communication network. The terminals may be mobile communication terminals, fixed information terminals, or robot terminals with display and input functions. Users view the information output by the server and perform necessary operations through the terminals.

[0219] In this invention, the server implements various functional modules through programs stored in the storage medium. When executed, these programs cause the server to perform specific data processing and operations. The server uses a scripting language runtime environment as its primary programming environment and leverages data analysis libraries (e.g., tabular data processing libraries), machine learning libraries (e.g., general machine learning algorithm libraries), deep learning frameworks (e.g., common tensor operation and neural network frameworks), visualization libraries (e.g., two-dimensional plotting libraries or interactive chart libraries), and network communication libraries to collaboratively complete a series of technical processes, including data acquisition, data fusion, model training, online inference, indicator calculation, prompt generation, generative artificial intelligence model invocation, and document generation.

[0220] In terms of data management, the server stores publicly available information from public institutions and information held by private entities in different logical tables within the database. For example, the server can create tables for public event records, monitoring device characteristics, population statistics, funding allocation, forecast results, policy indicators, and generated documents. At the storage layer, the server employs carefully designed column types and index structures, such as creating composite indexes for time and location fields, thereby reducing disk accesses and improving query performance when retrieving data by spatiotemporal region. This spatiotemporal-dimensional index configuration represents an optimization at the internal data structure level of the computer, distinct from traditional business systems that only query by identifier, thus maintaining low-latency responses even when dealing with large-scale, multi-source data.

[0221] During data fusion, the server uses a data analysis library to join and reconstruct multiple data tables read from the database. The server uses a two-dimensional table data structure (e.g., a row-column data frame) as a unified intermediate representation in memory. Based on time information, the server normalizes records to a uniform time granularity (e.g., rounding all timestamps down to 30-minute or 1-hour granularity), and maps place names and addresses from different sources to a unified regional code or geographic grid number based on location information. The server integrates public institution event records with civilian surveillance features, demographics, and financial information into a single data frame through key-value joins based on "time granularity × regional code." During this process, the server performs preprocessing operations such as missing value imputation, outlier truncation, and category encoding to form a feature matrix suitable for both batch training and real-time inference.

[0222] When building a machine learning model, the server takes feature vectors constructed from fused data and target labels as input. The server can employ various algorithms, such as tree models, linear models, or neural network models. As a preferred approach, the server can build a gradient boosting tree-based classification model to estimate the probability of a specific event occurring in a given spatiotemporal region within a future time window. Features selected by the server in this model may include: recent pedestrian traffic statistics, the frequency of past events, regional population density, existing surveillance device density, historical investment, and other environmental characteristics. The server uses a loss function (e.g., logarithmic loss) and a gradient boosting algorithm to iteratively train and update the model parameters. In each iteration, the server calculates the error between the current model's predicted output and the true label, adjusting the node partitions and leaf node weights in each basis function tree based on gradient information, thereby obtaining a model with superior performance on the validation set after a finite number of iterations.

[0223] In another implementation, the server can employ a neural network model to represent more complex spatiotemporal dependencies. The server can construct a multi-layer feedforward neural network or a sequence-based neural network to perform non-linear mapping of input features. In this architecture, the server sets several hidden layers, each containing several neurons, and uses rectified linear units or other non-linear activation functions. The server updates the weight matrix and bias vector by minimizing the cross-entropy loss function, using backpropagation and gradient descent variants (such as adaptive learning rate optimization algorithms). During training, the server can use batch normalization, L2 regularization, and early stopping strategies to reduce overfitting and improve model generalization ability and prediction stability.

[0224] In this invention, the server employs a feature organization method based on spatiotemporal regions as the basic unit, meaning each training sample corresponds to a specific time period and a specific spatial region. This data structure allows the server to finely characterize the risk levels of different regions at different times. Simultaneously, the server models short-term trends by constructing historical features for each region based on time windows, such as event counts and pedestrian flow sequences for the most recent N time periods. Compared to traditional methods that only use static regional attributes, the server significantly improves the predictive sensitivity for sudden risks by introducing sequential features, representing a substantial improvement to the prediction algorithm structure.

[0225] After completing model training, the server serializes and stores the trained model object and its matching auxiliary components such as normalizers and encoders in non-volatile storage. During the subsequent online inference phase, the server loads these components, transforms newly arriving real-time data into feature vectors through a processing pipeline identical to that used in the training phase, and then inputs these vectors into the model to calculate the probability of events occurring. The server compares these probabilities with set thresholds to assign a risk level to each spatiotemporal region. By aggregating the number of high-risk areas, the population covered, and the status of existing surveillance devices, the server calculates patrol behavior indicators and surveillance device configuration indicators. For example, the server can generate a structured countermeasure indicator table based on rules such as "for each additional high-risk area, the number of patrols increases by a certain baseline value" or "when the existing surveillance device density in a certain area is lower than the target value, calculate the number of new devices required." These rules are implemented internally as deterministic algorithmic logic, unlike traditional human experience-based judgments, and can be uniformly executed across large-scale areas.

[0226] In calculating funding allocation indicators, the server integrates predicted event risks, population density, and existing infrastructure conditions into a unified score. The server can use linear weighted models or multi-objective optimization methods to unify indicators from different dimensions into a funding allocation priority ranking. The server uses budget constraints as optimization conditions to calculate recommended funding allocations for each region. In this way, the server not only outputs a single risk prediction result but also maps multi-dimensional factors into actionable resource scheduling suggestions, achieving a technical chain that accurately connects data analysis to countermeasure indicators.

[0227] When generating prompts and invoking generative AI models, the server first organizes the strategy indicators and evaluation benchmarks into structured text descriptions or semi-structured information. The prompts constructed by the server include a task description section and a data description section. For example, the server can generate the following prompts: "Based on the following forecast data and crime prevention countermeasures indicators, please write a brief description for an internal report to the municipality, including: an overview of high-risk areas, recommended patrol and camera installation measures, and expected results. Please use a formal tone and keep the length approximately 500 words. The data is as follows:..."

[0228] For example, the server can generate the following prompt: "Please use crime data from the past year and burglar alarm camera analysis data to predict crime rates for the next week, and set crime prevention KPIs for each street (e.g., number of daily patrols, number of new cameras needed). Please also provide a brief written description so that municipal staff can directly reference it in their reports."

[0229] The server concatenates the aforementioned prompts with the selected structured data to form an input sequence, which is then sent to the inference service where the generative AI model resides via a network interface. In one implementation, the generative AI model can be an autoregressive language model based on the Transformer architecture, featuring a multi-layered self-attention encoding and decoding structure. This model is pre-trained on a large-scale corpus and fine-tuned on domain-specific texts to achieve strong professional text generation capabilities. The server limits the length, style, and paragraph structure of the output text through control information embedded in the prompts, ensuring that the generated results meet the requirements of reports and application documents in both format and content. This method of jointly controlling the generation process through structured data and prompts significantly improves the consistency between the output content and the underlying data compared to the traditional method of only inputting natural language instructions, reducing the burden of human proofreading and representing a technical improvement to the way generative AI models are invoked.

[0230] After receiving the natural language text output by the generative artificial intelligence model, the server associates the text with pre-generated statistical charts. The server can use a plotting library to generate risk heatmaps, trend line charts, and bar charts divided by region, writing the image file identifiers along with the text content into a document information table. The server can further combine the text and charts into a structured document format, such as a document structure recognizable by readers or editors, allowing the terminal to present summaries, detailed analyses, countermeasure lists, and chart descriptions in segments. This document generation process is implemented internally by the server through a series of data structure transformations and formatting operations. Compared to traditional manual typesetting methods, this significantly reduces the number of I / O operations and repetitive format conversions, thus improving the overall generation speed.

[0231] In this invention, the terminal serves as both the user interface presenter and the input collector. The terminal communicates with the server via a network, and the application running on the terminal receives notification information and document summaries from the server. The terminal displays this content in a graphical user interface, allowing users to view detailed reports and countermeasure indicators. Upon receiving a structured notification from the server, the terminal can mark high-risk areas in a map component with different colors based on field information, and display area names and recommended patrol times in a list component, sorted by risk level. The terminal can also read some information aloud to the user via a speech synthesis module to enhance accessibility. After the user completes a patrol or implements countermeasures, the terminal collects the implementation results information through the input interface and sends it to the server.

[0232] Users browse the countermeasure indicators and document information provided by the server through the terminal interface and formulate specific action plans accordingly. Users can arrange personnel shifts based on the server's suggested patrol behavior indicators and develop equipment procurement and installation plans based on monitoring device configuration indicators. During execution, users record actual progress and submit feedback information through the terminal. Users do not need to understand the details of the server's internal algorithms to utilize the system's functions, but the server continuously adjusts its models and parameter configurations based on user feedback, forming an adaptive improvement mechanism.

[0233] After receiving the implementation results, the server writes this feedback data into an implementation record table and compares it with the actual events occurring in subsequent time periods. The server can update the machine learning model through incremental learning or periodic retraining. For example, during retraining, the server adds recent implementation results as new features to the training dataset, allowing the model to learn the impact of policy implementation on the probability of events. The server introduces a weight adjustment mechanism related to implementation effectiveness into the error function, making the model focus more on policy-sensitive areas during prediction. This closed-loop feedback-based model update method enables the system to continuously improve prediction accuracy and resource allocation rationality over long-term operation, demonstrating an enhancement of the reasoning and adaptability of computer technology itself.

[0234] In terms of communication load control, the server reduces the number of calls to the generative AI model by pre-compiling and caching frequently accessed query results and general strategy templates locally. When generating large-scale reports, the server can use batch processing to integrate structured data from multiple regions into a single request. By embedding multi-region data tags in the prompt, the server guides the generative AI model to output data in segments by region, thereby reducing network round trips and model startup overhead. This optimization of the generative AI model's calling pattern technically reduces average response time and network bandwidth usage, representing a performance improvement in computer systems that goes beyond traditional document generation business logic.

[0235] Servers can employ different algorithms and architectures in various implementations. For example, in one implementation, the server can use temporal convolutional networks or graph neural networks to process spatiotemporal features, treating regions as graph nodes and adjacency relationships between regions as graph edges. Risk information is propagated across the graph through convolutional operations to better capture the risk diffusion effect between neighboring regions. In another implementation, the server can use a self-attention mechanism to assign different weights to features at different time steps, allowing the model to automatically focus on the time slices most critical to the prediction, thereby improving prediction accuracy and robustness. The server can also choose single-machine multi-threaded or distributed training schemes based on different data scales and hardware resource conditions to process large-scale datasets in parallel, improving model training speed.

[0236] Through the aforementioned implementation, the server not only achieves the automatic acquisition and fusion of publicly available information from public institutions and information held by private entities, but also employs specialized data structures, indexing mechanisms, feature engineering strategies, and model architecture design internally to improve the efficiency and prediction accuracy of multi-source heterogeneous data processing. By generating structured prompts and controlling the output of generative artificial intelligence models, the server achieves controlled transformation from underlying data to high-level natural language documents, avoiding the problem of disconnect between generated content and data. Terminals and users primarily play the roles of information display and feedback within the system, while core data processing and intelligent reasoning are all completed internally by the server in the form of reproducible algorithms. Therefore, this invention is not simply business process automation, but rather achieves multiple technical effects such as increased computing speed, improved prediction accuracy, optimized data management, and reduced communication load through specialized design of the computer's internal data flow, model structure, and calling patterns.

[0237] use Figure 12 The processing procedure is explained.

[0238] Step 1: The server retrieves raw data from an external data source and writes it to the database.

[0239] Inputs: Open data files or interface responses from public institution open platforms, monitoring device information or processed video feature data from private entity systems.

[0240] The server sends requests to the open data platform through a communication interface to download files containing population statistics, event logs, etc., or obtains structured responses through the interface; at the same time, the server receives relevant data from monitoring devices in civilian systems through interfaces, file transfers, or message queues.

[0241] The server uses a data parsing program to parse data of different formats (such as tabular files, text files, and structured data returned by interfaces) into an internally unified record format, converts time and location fields into a standard format, and writes the parsed records in batches into the corresponding data tables in the database.

[0242] Output: Original public and private data records written to the database according to field definitions.

[0243] Step 2: The server aligns data from different sources based on time and location information and generates fused data.

[0244] Input: Original public data records and original private data records stored in the database.

[0245] The server reads event records, demographic records, and monitoring feature records from the database within a specified time range; the server standardizes the time field, unifying all time values ​​to a preset time granularity; the server standardizes addresses or place names, mapping them to unified regional codes or geographic grid numbers.

[0246] The server uses "time granularity + regional code" as the key to match and merge records from different sources, combining population information, behavioral information, event occurrence information and financial information in the same spatiotemporal unit into a single record, and constructing a two-dimensional table-like data structure in memory; then the server writes these aligned records as fused data into the fused data table in the database.

[0247] Output: A collection of fused data records containing multi-source attribute fields organized by spatiotemporal units.

[0248] Step 3: The server extracts attribute information from the fused data and generates datasets for learning and for inference.

[0249] Input: A collection of merged data records from the database.

[0250] The server reads records from the fused data table, including fields such as population information, behavioral information, event occurrence information, and financial information; the server performs field checks on each record, discards records with serious missing or obvious anomalies, and imputes some missing fields; the server scales or standardizes numerical fields and encodes categorical fields to obtain numerical feature vectors.

[0251] The server divides the processed records into a learning dataset and an inference dataset according to a predetermined time or proportional division method, and stores them in an in-memory data structure or a new database table, respectively.

[0252] Output: A dataset for learning containing feature vectors and target labels, and a dataset for inference containing feature vectors.

[0253] Step 4: The server builds machine learning models based on the dataset used for learning.

[0254] Input: Feature vectors and corresponding target labels from the learning dataset.

[0255] The server extracts the feature matrix as input and the target label as output from the learning dataset; the server selects the algorithm structure according to the configuration, such as a tree-based model or a neural network-based model, and initializes the model parameters or weights.

[0256] The server repeatedly performs forward computation and error calculation: for each batch of training samples, the server calculates the error between the model output and the true label, and adjusts the parameters according to the error; for tree models, the server selects split points in the feature dimension to construct a decision tree and updates the leaf node weights; for neural network models, the server calculates the gradient of the weights of each layer through the backpropagation algorithm, and updates the weight matrix and bias according to the set learning rate.

[0257] After multiple iterations, the server selects the optimal model parameter configuration based on the performance metrics of the validation set, and stores the trained model structure and parameters in a non-volatile storage medium.

[0258] Output: A trained machine learning model with optimized parameters.

[0259] Step 5: The server uses the inference dataset and real-time arriving data to perform probabilistic inference about the occurrence of events.

[0260] Input: Feature vectors from the dataset used for inference, and real-time feature vectors generated based on the latest records.

[0261] The server reads feature vectors from the inference dataset and real-time recordings; the server inputs these vectors into a trained machine learning model, performs forward computation, and obtains the probability value of the event occurrence for each sample.

[0262] The server compares the output probability with a pre-set threshold, marks each spatiotemporal unit with a risk level based on the probability, and records the prediction results, including the spatiotemporal identifier, the probability of the event, and the risk level, into the prediction result table.

[0263] Output: A set of prediction results records with the probability of the event occurring and the risk level.

[0264] Step 6: The server calculates policy indicators and evaluation benchmarks based on the prediction results.

[0265] Inputs: a set of prediction result records, population information from the fused data, existing monitoring configuration information, and funding information.

[0266] Based on the probability of event occurrence and risk level of each spatiotemporal unit, combined with population size, existing monitoring device density, and historical funding, the server uses rule-based or optimization algorithms to calculate patrol behavior indicators, monitoring device configuration indicators, and funding allocation indicators. For example, the server calculates the number of patrols needed for each area based on risk level and population density, and calculates the number of new devices needed for each area based on the difference between the current number of monitoring devices and the target density.

[0267] The server summarizes the countermeasure indicators for all regions, further calculates evaluation benchmarks, such as the overall risk reduction target, the indicator value of fund utilization efficiency, and the degree of balance in resource allocation for each region, and writes the results into the countermeasure indicator table and the evaluation benchmark table.

[0268] Output: A set of countermeasure indicators including patrol behavior indicators, monitoring device configuration indicators, and fund allocation indicators, as well as corresponding evaluation benchmark records.

[0269] Step 7: The server converts the countermeasure indicators and evaluation benchmarks into structured data and generates prompt statements.

[0270] Input: Set of policy indicators and evaluation benchmark records.

[0271] The server selects the fields needed for document generation from the countermeasure indicator table and the evaluation benchmark table, constructs a key-value pair structure or field list according to the predetermined field order, and generates a machine-readable structured data representation; the server encodes this structured data into text fragments or embedded descriptions.

[0272] Based on the target document type and user attributes, the server generates a prompt statement containing task instructions, output format requirements, and data descriptions. For example, the server generates the prompt statement "Please write a description for an internal report of the municipality based on the following predicted data and crime prevention countermeasure indicators, including: an overview of high-risk areas, recommended patrol and camera installation measures, and expected results. Please use a formal tone and keep the length to approximately 500 words. The data is as follows:...", and appends structured data content to the "Data is as follows" section.

[0273] Output: Structured data text and corresponding prompts used as input to the generative artificial intelligence model.

[0274] Step 8: The server calls a generative artificial intelligence model to generate document information.

[0275] Input: Prompt statements and structured data text constructed by the server.

[0276] The server sends a request to the inference service where the generative AI model resides via a network interface, using the prompt and structured data as input to the model. The generative AI model encodes the input sequence based on its internal neural network structure and generates the corresponding natural language output text.

[0277] The server receives natural language text returned by the generative artificial intelligence model, parses it, and segments it according to document type, such as extracting the overview section, detailed analysis section, and countermeasure description section. If necessary, it inserts chart placeholders or field reference marks into the text and writes the sorted text along with metadata into the document information table.

[0278] Output: Document information records including report information, application information, and supplementary information.

[0279] Step 9: The server generates a notification message and sends it to the terminal.

[0280] Input: Set of countermeasure indicators, document information records, and user or terminal identification information.

[0281] The server filters out high-risk areas and key measures that require immediate notification based on countermeasure indicators, and compresses them into short text notification content, such as including area name, time period, risk level, and recommended number of patrols; the server selects an appropriate notification format based on user or terminal type, such as plain text notification or notification with a link.

[0282] The server encapsulates the notification content into a message structure required by the communication protocol, attaches an identifier or link to detailed document information in the message, and sends the notification information to the target terminal over the network.

[0283] Output: Notification messages sent to each terminal.

[0284] Step 10: The terminal receives notification information from the server and displays it to the user.

[0285] Input: Notification messages and document information identifiers sent by the server.

[0286] The terminal receives notification messages through the communication module and parses the content and identification fields in the messages. The terminal displays the area name, risk level, and brief countermeasures information to the user in a list or reminder format on the display interface, and requests the corresponding document information from the server when the user selects to view details.

[0287] After receiving the detailed document content, the terminal will arrange and display the text and charts on the interface. For example, it will display high-risk areas on the map view and explanatory paragraphs output by the generative artificial intelligence model in the text view.

[0288] Output: Prevention measures information and detailed descriptions displayed on the terminal interface.

[0289] Step 11: Users implement countermeasures based on the information displayed on the terminal and provide feedback on the results.

[0290] Input: Countermeasure indicators and document information displayed on the terminal.

[0291] Users read the high-risk areas and suggested measures displayed on the terminal, arrange actual patrol activities according to patrol behavior indicators, arrange equipment installation or adjustment according to monitoring device configuration indicators, and formulate a fund usage plan according to fund allocation indicators. After implementation, users fill in the actual number of patrols completed, the number of new devices added, and other implementation details through the terminal's input interface.

[0292] The user confirms the input and submits feedback through the terminal. The terminal then packages the implementation result information into a structured record and sends it to the server.

[0293] Output: Records of implementation results uploaded via the terminal.

[0294] Step 12: The server receives the implementation results and updates the data and model.

[0295] Input: Records of implementation result information sent by the terminal and records of actual events occurring in subsequent time periods.

[0296] The server writes the implementation result information into the implementation record table and associates it with the actual event occurrence records of the corresponding region and time period. Based on the association results, the server calculates the differences in the occurrence of events before and after the implementation of the countermeasures and incorporates these differences as new features or labels into the subsequent training data.

[0297] The server loads historical fusion data and new implementation result records at predetermined times, reconstructs the expanded learning dataset, retrains or incrementally updates existing machine learning models, and adjusts model parameters to make the model more sensitive to implementation effects. At the same time, the server adjusts the prompt statement template and countermeasure indicator calculation rules according to the new model performance and the latest data, thus forming an automatic improvement mechanism based on feedback.

[0298] Output: Updated fused data records, updated machine learning models, and adjusted prompts and policy indicator calculation configurations.

[0299] Alternatively, an emotion engine for inferring user emotions can be combined. That is, the specific processing unit 290 can also use the emotion-specific model 59 to infer user emotions and perform specific processing using user emotions.

[0300] Example 2 The flow of a specific process in Example 2 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. The data processing device 12 will be referred to as the "server," and the smart device 14 as the "terminal."

[0301] Existing computer-generated document technologies suffer from the following problems. First, servers typically use only a single data source from public or private information as input, lacking the ability to automatically integrate and process multi-source heterogeneous data for consistency. This results in an incomplete and inconsistent data foundation for generating reports or application documents, thus affecting the accuracy and credibility of the document content. Second, traditional systems often rely on fixed rule templates, simply filling structured data into preset fields. They lack a unified processing flow for in-depth data analysis, evaluation indicator calculation, and chart visualization generation. This necessitates multiple manual configurations or the splicing of multiple toolchains on the server side, leading to high system complexity, poor maintainability, and low operating efficiency.

[0302] Furthermore, even if some systems introduce generative artificial intelligence models to generate natural language explanatory text, the prompts are often manually edited on the spot. The server lacks a mechanism to automatically construct prompts based on structured data, and also lacks a technical solution to automatically compare and verify the consistency between the model output text and the underlying data set. As a result, key information such as time, region, and quantity in the generated text may not match the actual data, increasing the burden of manual review.

[0303] In addition, when existing technologies update server-generated documents using terminal editing results, they usually only replace the text content without performing a programmatic consistency check on the updated document content against the original integrated data set or automatically identifying abnormal elements on the server side. This makes it impossible for the system to detect and prompt data logic conflicts or filling errors in a timely manner at the computer level, making it difficult to achieve end-to-end quality control of the entire document generation process.

[0304] In summary, designing an integrated processing mechanism on the server side that integrates multi-source data, calculates evaluation indicators, visualizes charts, links generative artificial intelligence models, and verifies consistency between terminal editing processes, thereby generating electronic documents for reports and applications with a higher degree of automation and greater consistency control within the computer, has become a pressing technical challenge in this field.

[0305] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Embodiment 2 is achieved by the following means.

[0306] In this invention, the server includes: a device for communicating with multiple information sources containing public and private information, obtaining the public and private information using structured queries, and standardizing the obtained information in computer memory to integrate it into a single data set; a device for performing missing data processing, type conversion, summary operations, and evaluation index calculations on the integrated data set using a general-purpose program execution environment and data parsing software to generate structured data corresponding to each item in the document template; and a device for automatically generating tables and graphs on the server side using visualization software based on the integrated data set and the evaluation indexes, and integrating the tables and graphs with predetermined values ​​within the document template. The system includes: a device for associating locations; a device for automatically constructing prompts for a generative artificial intelligence model on the server side based on the structured data and the document template, providing the prompts and a summary of the structured data to the generative artificial intelligence model to generate explanatory text, and arranging the explanatory text in a predetermined description area within the document template; and a device for generating predetermined electronic file formats for report documents and application documents based on the structured data, tables, graphics, and explanatory text arranged in the document template, saving the electronic files to an information storage device, and updating, verifying consistency, and identifying abnormal elements in the documents in conjunction with editing content received from the terminal. This allows for the automatic generation and updating of multi-source data-driven report documents and application documents through an integrated data integration, parsing, visualization, and generative artificial intelligence model processing flow within the server. It also improves the consistency control capabilities between data and text at the computer technology level, reduces manual intervention and review costs, and enhances the overall processing efficiency and reliability of the document generation system.

[0307] "Public information" refers to data and materials released to the public by public institutions, including but not limited to policy data, statistical data, financial data, project data and related metadata.

[0308] "Civilian information" refers to data and materials held or generated by non-public entities, including but not limited to business operation data, market research data, business transaction data, user behavior data and related metadata.

[0309] "Information source" refers to any data provider or data storage device that stores or provides public or private information, including local databases, remote databases, file storage systems, and network service interfaces.

[0310] "Structured query" refers to the operation method of retrieving data from information sources using a query language with fixed syntax rules, including but not limited to relational query language, object query language, and a set of instructions for filtering, sorting, and aggregating data.

[0311] A "single dataset" refers to a unified data view formed through data integration and processing within a server. This view normalizes, merges, and indexes heterogeneous data from multiple information sources according to a predetermined data model, so as to enable unified analysis and processing.

[0312] A "general program execution environment" refers to a computing environment that can run various program codes to complete data processing tasks, including but not limited to operating system environments, script interpretation environments, and application runtime environments.

[0313] "Data parsing software" refers to programs or libraries used to read, clean, transform, and analyze data, including but not limited to functional modules that implement missing value handling, type conversion, aggregation operations, and statistical calculations.

[0314] "Missing handling" refers to the operations performed on missing or incomplete data items in a dataset, such as identification, deletion, filling, or replacement, to improve the integrity and usability of the dataset.

[0315] "Type conversion" refers to the process of converting data from one data type to another, including but not limited to conversion between numeric, text, and time types.

[0316] "Summary operations" refer to operations that perform grouping, statistical and aggregation calculations on a data set, including but not limited to summation, counting, average, extreme values ​​and grouping statistics.

[0317] "Evaluation indicators" refer to quantitative or qualitative parameters used to measure the effectiveness, progress, or quality of a specific object, project, or activity. Their values ​​are obtained through calculation based on a unified dataset.

[0318] A "document template" is a pre-defined structured document framework for generating electronic documents, which includes placeholder areas for inserting text, tables, graphics, and other content, along with their formatting definitions.

[0319] "Structured data" refers to data organized according to a predetermined data model and having clearly defined field names and data types, including but not limited to tabular data, key-value pair data, and record sets.

[0320] "Visualization software" refers to programs or libraries used to generate graphics, tables or other forms of visualization based on structured data, including but not limited to drawing engines, chart generation modules and report generation components.

[0321] A "table" is a visual structure that presents data in rows and columns, used to show the correspondence between fields and records and statistical results.

[0322] "Graphics" refers to visual elements used to express data distribution, trends, or comparative relationships, including but not limited to bar charts, line charts, pie charts, radar charts, and combinations thereof.

[0323] "Generative artificial intelligence model" refers to a computational model that is based on machine learning or deep learning technology and can automatically generate text, images or other data content according to input information. In this invention, it specifically refers to a text generation model used to generate explanatory text.

[0324] "Prompt statements" refer to the instructional or descriptive text input into a generative artificial intelligence model to guide it in generating target content. These statements may include task descriptions, tone requirements, length requirements, and background information related to the data.

[0325] "Explanatory text" refers to natural language text output by a generative artificial intelligence model or processed based on its output, used to describe and explain data, evaluation metrics, or project status in a document.

[0326] "Pre-defined description area" refers to a specific location or field in a document template that is pre-defined for inserting explanatory text or other text content, including paragraphs, chapters, or text boxes.

[0327] "Documents used for reporting" refers to electronic documents generated for the purpose of presenting analysis results, business status, or project progress, whose structure and content conform to the predetermined report format requirements.

[0328] "Documents used for application" refers to electronic documents generated for the purpose of applying, submitting, or requesting approval, whose structure and content conform to the predetermined application or submission format requirements.

[0329] "Electronic document format" refers to the digital encoding format used to store and exchange document content in computer systems, including but not limited to text format, layout document format, and editable document format.

[0330] "Information storage device" refers to a hardware or software system used to preserve electronic documents and related data in digital form for a long or short period of time, including local storage devices and remote storage services.

[0331] "Terminal device" refers to a computing device operated by a user to access server functions and browse or edit document content, including but not limited to computer terminals, mobile terminals and embedded interactive devices.

[0332] "User-edited content" refers to information added, modified, or deleted by users through terminal devices and sent to the server regarding document content or related data.

[0333] "Consistency check" refers to the process by which a server compares the data or text content in a document with the unified data set to determine whether they maintain logical consistency and data consistency in terms of values, time, region, etc.

[0334] “Abnormal elements” refer to document content elements that are identified as inconsistent, erroneous, or suspected of being abnormal with the unified dataset during the consistency check process, including field values, descriptive text, or statistical results.

[0335] "Warning message" refers to the message content generated by the server after detecting abnormal elements to alert the user. It may include the location of the abnormality, the type of abnormality, and suggested handling methods.

[0336] In one embodiment of the present invention, the server includes a processor, main memory, a network interface, and a non-volatile storage device. The server runs applications within an operating system environment, where the operating system can be a general-purpose server operating system, such as a Unix-like operating system. The server installs database management software, data parsing software, and data visualization software in this environment, such as a relational database management system, a data analysis library, and a graphics generation library. The server also communicates with terminals and multiple information sources via the network interface, including databases or network service interfaces that store public and private information.

[0337] The server stores program instructions for implementing the functions of this invention in a non-volatile storage device. These programs are loaded into main memory upon server startup and executed by the processor. During program initialization, the server establishes a database connection pool, loads document template files, loads generative artificial intelligence model access configurations, and initializes the logging and caching modules. The server stores multiple document templates in the storage device; these templates can be in editable document formats or markup language formats and include predetermined description areas for inserting structured data, tables, graphics, and explanatory text.

[0338] When retrieving data, the server establishes sessions with multiple information sources. Using a database driver or network protocol client, the server sends query requests to the information sources using Structured Query Language (SCL) to retrieve records related to public and private information. The server loads the data returned by different information sources into an in-memory data structure, such as a two-dimensional array in tabular form or a list of records. The server maps and standardizes data fields from different information sources, unifying field names and data types to a predefined data schema, thereby constructing a single data set in memory. The server aggregates information originally scattered across multiple tables or records into a unified record structure by performing join operations on identifier fields (such as region codes, years, and project numbers).

[0339] During the data parsing phase, the server uses data parsing software to perform missing value handling, type conversion, and aggregation operations on a single dataset. The server fills in fields with missing values ​​using predefined rules, which may include using historical averages, medians, or default values ​​specified by business rules. The server converts text-type numeric fields to numeric types, converts date strings to internal timestamp formats, and maps categorical fields to discrete codes for subsequent statistical analysis and modeling. Through grouping and aggregation operations, the server summarizes data by year, region, project type, and other dimensions, calculating totals, averages, maximums, minimums, and counts to obtain the basic statistics used to calculate evaluation indicators.

[0340] In calculating evaluation indicators, the server utilizes functions defined in the general program execution environment to perform operations such as weighted summation, proportional calculation, normalization, and score mapping on the summarized results. The server can maintain a set of parameters for each evaluation indicator, such as weight coefficients, thresholds, and score intervals. Based on these parameters, the server maps the raw numerical values ​​to dimensionless score values ​​and stores the scoring results in structured data records. The server can also select different indicator weighting schemes based on different project categories or regional characteristics, thereby achieving differentiated technical configurations during the calculation process.

[0341] When generating visualizations, the server invokes data visualization software to map structured data into tables and graphs. The server constructs the table description structure in memory, specifying column headers, data rows, and formatting attributes, and can generate an intermediate rendering representation of the table, such as a table structure in markup language. When drawing graphs, the server uses a specific evaluation indicator or statistic as the vertical axis and time or category as the horizontal axis to generate bar charts or line charts; or it uses the proportion of each category as input to generate pie charts. The server sets titles, axis labels, scale ranges, and legends for the graphs and renders them as bitmap or vector graphics files. The server records the file path and corresponding insertion position identifier for each graph for subsequent use by the document generation module.

[0342] After generating structured data and visualization results, the server begins constructing the prompts required by the generative AI model. The server extracts key fields from the structured data, such as project name, target year, region name, main evaluation indicators and their values, and generates explanatory text through string concatenation or template filling. This content is then combined into a single prompt. The server can use different prompt templates depending on the target document section (e.g., "Project Background and Necessity" or "Implementation Results and Achievement of Evaluation Indicators") to control the theme and style of the generated text. Simultaneously, the server generates a data summary, organizing key numerical information into concise text descriptions to assist the model in generating content that better reflects the data.

[0343] The server can invoke the generative artificial intelligence model using the following example prompt: "Based on the 2023 public information and enterprise project performance data of a certain region in the database, please write the 'Project Implementation Effect and Evaluation Indicator Achievement' section of the payment application, approximately 800 words in length, in a formal language suitable for submission to government departments. Key indicators include: increase in employment, sales growth rate, number of participating enterprises, and social impact evaluation score. Please provide an analysis and explanation based on these indicators." When a server invokes a generative AI model, it can send a request to a remote inference service via a network interface. The server includes a prompt and a data summary in the request, specifying model parameters such as the maximum length of the output text, temperature parameters, and the number of outputs. After receiving the explanatory text returned by the model, the server performs basic text normalization, such as removing extra spaces, standardizing punctuation, and restricting consecutive repetitive segments.

[0344] In one specific implementation, the server uses a sequence generation model based on a multi-layer attention mechanism as the generative artificial intelligence model. This model can employ a neural network structure containing multiple encoders and decoders, with each layer including a multi-head self-attention sublayer and a feedforward network sublayer. During model training, the server performs supervised learning using a training sample set consisting of historical report documents and corresponding data summaries. The server uses the cross-entropy loss function as the objective function during training and updates the network weights through backpropagation. The server can employ a variant of stochastic gradient descent optimization and use mini-batch training to improve training efficiency. The server can also perform data augmentation during training, such as generating multiple different explanatory texts for the same data summary, to improve the model's robustness to different prompts.

[0345] To improve consistency and accuracy, the server automatically validates the explanatory text generated by the model during runtime. The server extracts time, region, and quantity information from the explanatory text using rule matching or a lightweight parser, and compares this information with corresponding fields in a single dataset. The server sets tolerance rules for each information type; for example, monetary fields are allowed to fluctuate within a certain error range, while year fields must match exactly. Based on the comparison results, the server generates a consistency flag for each information unit. If inconsistencies or suspected errors are detected, the server marks that section as requiring review when generating the document and appends a warning message to the data structure returned to the terminal. This automatic validation process can detect potential errors internally, reducing the burden of subsequent manual review.

[0346] During the document generation phase, the server integrates structured data, tables, graphics, and explanatory text into the document template. The server iterates through placeholders in the template, inserting corresponding field values ​​or content into predefined description areas. The server performs formatting operations on text fields, including number formatting, date formatting, and unit display methods. The server inserts tables and graphics into the document object in the order specified in the template, controlling their size and layout to ensure readability and aesthetic appeal in the output file. After rendering, the server converts the document into a layout-based or editable document according to the user-selected output format, stores the generated file in an information storage device, and records file metadata.

[0347] In this invention, the terminal serves as a device for user interaction with the server. The terminal displays a user interface provided by the server through a browser or dedicated application. The terminal receives parameter input from the user, such as the target year, region, project category, required document type, and whether to enable generative artificial intelligence model-assisted writing. The terminal packages these parameters into a request message and sends it to the server via the network. After receiving document preview data returned by the server, the terminal renders the document content into a visual interface, displaying text, tables, graphics, and warning labels related to each section of the content.

[0348] Users can browse automatically generated document content on the terminal interface and modify explanatory text or certain fields through the text editing area. Users can adjust the tone of project descriptions, add specific background information, or correct suspected errors detected by the system. After users submit their modifications on the interface, the terminal sends the modified text or field values ​​to the server. The terminal can also allow users to directly edit prompts on the interface to re-invoke the generative AI model to generate alternative text, thereby enabling collaborative editing of content generated by humans and the model.

[0349] After receiving user-edited content, the server replaces or merges this content with the corresponding parts in the original document. The server then performs a consistency check again, comparing the user-modified data or text with a single dataset and generating a new consistency report. For content that passes the check, the server saves it as the final document content; for elements that still have inconsistencies, the server logs detailed information internally to adjust system configurations and rules. The server then re-renders the finalized document into the target file format, updates the file in the information storage device, and provides the user with a download link or access portal via the terminal.

[0350] In this invention, the server achieves several technical improvements through the aforementioned modular processing flow. First, at the data level, the server reduces repeated queries and data transformations by constructing a unified data model and a single data set, thereby reducing database access frequency and network communication load, and improving data processing speed. Second, during the calculation of evaluation metrics and the generation of visualizations, the server reuses the same in-memory data structure multiple times, avoiding unnecessary data copying, improving memory utilization efficiency, and reducing the overhead of intermediate data serialization and deserialization. Third, by automatically constructing prompt statements and performing automatic consistency checks, the server establishes a feedback loop between the output of the generative artificial intelligence model and the underlying data, enabling technical correction of the model output rather than simply relying on manual post-processing.

[0351] During the training and inference processes of generative AI models, the server selects appropriate loss functions and optimization algorithms, and structures the training data to ensure that the update direction of the model's internal parameters more directly reflects the data consistency goal. This reduces the probability of inconsistency between the model output and the data, thereby improving the overall accuracy of the output text. On the inference side, the server can dynamically adjust model parameters, such as generation temperature and maximum output length, based on data complexity and document length. This reduces the generation of irrelevant content while maintaining readability, thus improving inference efficiency and reducing computational resource consumption.

[0352] This invention also considers variations of different implementations. In one variation, the server can deploy the generative artificial intelligence model on a local hardware acceleration device, such as a graphics processing unit or a dedicated inference accelerator, to further reduce network latency and improve inference speed. In another variation, the server can distribute the data parsing module and visualization module across multiple computing nodes, processing them in parallel through task queues to handle large-scale data scenarios. In yet another variation, the server can use a rule engine to configure the logic for constructing prompt statements, allowing different organizations or business scenarios to customize the structure of prompt statements through configuration without modifying the program code, thereby improving the scalability and maintainability of the system.

[0353] Through the specific structure and processing flow described above, this invention organically combines multi-source data integration, evaluation index calculation, chart visualization, and the text generation capabilities of generative artificial intelligence models to form a highly automated document generation technology solution with self-checking capabilities within a server. This solution implements technical features such as data structure optimization, algorithm process improvement, and model output constraints within the computer. Therefore, it is not merely a simple automation of manually written reports, but rather a substantial improvement over existing computer technology in terms of processing speed, data and text consistency, resource utilization efficiency, and error rate control.

[0354] use Figure 13 The processing procedure is explained.

[0355] Step 1: The server receives a generation request from the terminal. The input is the request message sent by the terminal, which includes at least the target year, region identifier, project identifier, document type (report or application), whether generative AI models are enabled, and user identity information. The server parses the request, checks the completeness and format validity of the parameters, and creates a task object in memory. The server performs permission verification based on the user's identity information; if successful, the task object is written to the log system and proceeds to further processing. The output is a task object containing a set of normalized parameters.

[0356] Step 2: The server accesses information sources to obtain raw data. The input is the task object generated in step 1, which includes database connection configuration and query conditions (year, region, project type, etc.). The server sends structured query instructions to at least one public information data source and at least one private information data source via a database driver or network interface. The server receives the record sets returned by each information source and loads them into an in-memory data structure, such as a two-dimensional array of multiple tables. The server does not modify the data content in this step; it only performs data extraction and preliminary classification. The output is a set of raw data categorized by information source.

[0357] Step 3: The server performs data consolidation and standardization. The input consists of multiple raw datasets obtained in step 2. First, the server maps field names from different information sources to unified field names and converts field types to a unified data type, based on a predefined data schema; for example, it unifies "region code" and "region ID" from different data sources into the internal field "region_code". The server uses join keys (such as region_code, year, project_id) to perform join operations on records from different sources, achieving record-level merging. The server applies priority rules to conflicting fields, retaining values ​​from data sources with higher credibility. Finally, the server constructs a single dataset in memory, storing all related fields in a tabular structure. The output is a single dataset with unified field definitions.

[0358] Step 4: The server performs missing data handling and type conversion. The input is a single dataset generated in step 3. The server scans each field, identifying cells with null or invalid values. The server performs imputation operations according to the missing data handling strategy configured for each field, such as imputing numeric fields with historical averages, imputing categorical fields with specified default categories, or marking records as low-confidence records when there are severe missing values. The server also performs numeric type conversion on string-based numbers, parses date strings into internal timestamp format, and normalizes Boolean expressions into a uniform truth value representation. The server updates data quality metrics such as missing rate and conversion error count during processing. The output is a cleaned and type-normalized dataset.

[0359] Step 5: The server performs aggregation calculations and evaluation index calculations. The input is the dataset output from step 4. Based on predefined aggregation rules, the server groups the data by dimensions such as year, region, and project category. For each group, the server calculates statistics such as total amount, average, maximum, minimum, and count. Subsequently, the server uses these statistics and pre-configured weight parameters to perform evaluation index formula calculations, such as calculating the "employment contribution index," "economic growth index," and "social impact score." The server normalizes each index result, mapping it to a fixed range (e.g., 0 to 100), and may assign level labels based on the scoring range. The server appends the calculation results to the dataset, forming an extended dataset containing the original fields and evaluation index fields. The output is a structured dataset containing the evaluation indicators.

[0360] Step 6: The server generates a tabular data structure. The input consists of the extended data set obtained in step 5 and the target document type information. Based on a predefined table template configuration, the server selects the fields to be displayed and their column order, such as year, budget amount, actual expenditure, and evaluation indicator scores. The server constructs a table description object in memory, setting the column name, data type, and display format (e.g., currency format, decimal places) for each column. The server iterates through the data records, mapping each record to a table row, generating a logical tabular data structure. If the number of data rows exceeds a preset threshold, the server can execute a pagination strategy or a sampling strategy to control the table size. The output is one or more tabular data structures for subsequent rendering.

[0361] Step 7: The server generates graphical visualization objects. The input is the extended data set and chart configuration parameters from step 5. Based on the configuration, the server selects the indicators to be visualized, such as a comparison of annual subsidy applications and actual expenditures, or the distribution of evaluation indicator scores across different regions. The server constructs a graphical data series, mapping the selected fields to horizontal and vertical axis data. The server sets the chart type (bar chart, line chart, pie chart, etc.), color scheme, axis labels, title, and legend. The server renders this data using visualization software to obtain graphical objects and exports them as image files (such as PNG or SVG). The server records the storage path and corresponding insertion position identifier for each image file. The output is a collection of graphical files and their metadata description.

[0362] Step 8: The server constructs prompts for the generative AI model. The input is the extended dataset from step 5 and the generation target specified in the task object (e.g., "Project Implementation Effect Description"). The server extracts key data items from the extended dataset, including year, region, project name, main evaluation indicators and their values. Based on a predefined prompt template, the server fills these data items into template placeholders to generate a natural language prompt. The server also generates a data summary text, such as a brief description combining key indicators and values. The server encapsulates the prompt and data summary into a model input parameter structure for subsequent calls to the generative AI model. The output is one or more text-based prompts and their corresponding data summaries.

[0363] Step 9: The server invokes a generative AI model to generate explanatory text. The input consists of the prompts and data summaries generated in step 8, as well as model call parameters (such as maximum output length and temperature coefficient). The server sends a request to the model server via a network interface, containing the prompts and data summaries. The model performs sequence generation computation on remote computing resources, outputting natural language explanatory text. After receiving the explanatory text returned by the model, the server performs basic post-processing: removing control characters, standardizing punctuation, and limiting excessively long paragraphs. The server establishes an association between the processed explanatory text and the corresponding prompts and data summaries, storing them in a memory structure. The output is a set of explanatory texts corresponding to a specific document section.

[0364] Step 10: The server performs a consistency check on the explanatory text. The input consists of the explanatory text output from step 9 and the extended data set from step 5. The server extracts time information (year, time period), regional information (region name, code), and quantity information (amount, number of people, ratio, etc.) from the explanatory text using rule matching or a lightweight parser. The server compares this extracted information item by item with the corresponding fields in the extended data set, employing predefined matching rules and tolerance settings. For example, the amount field needs to match within a certain error range, and the year field must be completely consistent. Based on the comparison results, the server marks each information unit as "pass," "suspicious," or "error," and generates a consistency result structure. The server records the text position of paragraphs marked as "error," providing a basis for subsequent user review prompts or triggering regeneration. The output consists of the explanatory text with consistency markers and the consistency result structure.

[0365] Step 11: The server binds structured data, tables, graphics, and explanatory text to a document template. The input consists of the table and graphic data structures from steps 6 and 7, the consistency marker explanatory text from step 10, and a pre-stored document template. The server parses the document template, identifying placeholders and predefined descriptive areas. The server inserts structured fields into corresponding text placeholders, maps table data structures to table elements in the document, inserts graphic file paths as image elements, and fills specific paragraph areas with explanatory text. The server specially marks problematic text based on consistency markers, for example, by adding marker attributes to the internal structure for highlighting during terminal rendering. The output is a complete document object model built in memory.

[0366] Step 12: The server generates and saves the target electronic file. The input consists of the document object model constructed in step 11 and the output format specified in the task object. The server selects the appropriate rendering module based on the output format: if it's a layout document format, the document object is converted into a page description structure and rendered as an electronic file; if it's an editable document format, the corresponding structure is generated and written to a file container. The server writes the generated electronic file to a specified directory on the information storage device, and simultaneously records metadata such as file path, generation time, user ID, and document type in the database. The server generates a download address or file identifier for the terminal to access and returns it as a response to the terminal. The output is the persistently stored electronic file and its access identifier.

[0367] Step 13: The terminal presents a document preview and warning messages to the user. Input consists of the document preview data returned by the server (which can be structured content or rendered results) and the consistency result structure. The terminal renders the document text, tables, and graphics in the display interface, while highlighting or marking suspicious or erroneous parts in the explanatory text with icons based on consistency markers. The terminal displays detailed warning messages in the interface sidebar or pop-up prompt area, explaining which data item is inconsistent and the possible reasons. The terminal provides editing controls, allowing users to modify relevant text or values ​​within the interface. Output is a visually presented document page and an interactive editing interface.

[0368] Step 14: Users edit document content or prompts on the terminal. Input consists of the document content and editing controls displayed on the terminal. Users modify descriptive text, add background information, or correct fields such as amounts and dates in the text input area. Users can also modify prompts used by generative AI models in dedicated input boxes, such as changing the required word count, tone, or emphasized metrics. After editing, users select "Save," "Regenerate," or "Generate Final Version" buttons on the interface. Output consists of the user-edited text content, field values, and any updated prompts; this content is packaged by the terminal into an update request and sent to the server.

[0369] Step 15: The server processes user-edited content and updates the document. The input is the update request sent by the terminal in step 14, which includes the modified text, field values, and potentially updated prompts. The server first writes the modified field values ​​back to the extended dataset or a separate user-modified structure, and then performs a consistency check again based on the updated data. If the user modifies the prompts and requests regeneration of the explanatory text, the server returns to the process of steps 8 and 9, re-invokes the generative AI model for the specified text segment, and can perform the consistency check in step 10 again. After confirming that the consistency between the data and the text meets the preset standard, the server updates the corresponding part in the document object model to the final content. The output is the updated document object model and the new electronic document version or the status of preparing to generate the final version of the document.

[0370] Step 16: The server generates the final electronic file and provides it for download. The input consists of the document object model updated in step 15 and the final output format selected by the user. The server invokes the rendering module to generate the final electronic file and saves it to an information storage device, while simultaneously generating a new file identifier and metadata record for this final version. The server sends a response message to the terminal containing the download address or file identifier of the final version file. Upon receiving this information, the terminal displays a "Download Final Version" button on the interface. The user can click this button to download the final file from the server for printing, archiving, or submission to external organizations. The output is a final electronic document usable by external systems or devices.

[0371] Application Example 2 The process flow corresponding to the specific processing in Use Case 2 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. In addition, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0372] Traditional data processing and document generation systems typically only perform simple statistics and fixed template filling on structured data, which presents the following technical problems: (1) In a multi-source heterogeneous data environment, it is difficult for computers to efficiently obtain structured and unstructured data from multiple public and civilian information sources and automatically integrate them according to a unified identifier. This results in a large amount of manual intervention, complex and error-prone data pipelines, which limits the ability of computer systems to automatically analyze and process large-scale datasets.

[0373] (2) When calculating evaluation indicators and generating reports, existing systems mostly use hard-coded rules and fixed output formats, lacking a general data abstraction and automatic modeling mechanism. They cannot flexibly define and adjust evaluation indicators according to different business scenarios (such as social infrastructure construction, funding provision, business evaluation, production management, etc.). The computer's reusability and scalability of evaluation logic are poor, which reduces the maintainability and portability of the entire system.

[0374] (3) At the user interaction level, existing technologies usually ignore the user's emotional state and generate documents or notices with uniform content based only on objective data. This makes it impossible for the computer system to dynamically adjust the evaluation goals and the complexity of the information to be presented according to the user's stress, anxiety or positive emotions. As a result, the user has a high understanding cost when facing complex data output, which is not conducive to improving the human-computer interaction experience.

[0375] (4) Existing natural language generation solutions often simply use generative artificial intelligence models as “text generators” without abstracting and controlling the construction of “prompt statements” at the system architecture level. They lack a technical means that allows computers to automatically generate prompt statements based on integrated data, evaluation indicators, emotional states and visualization results, making it difficult to stably produce structured, controllable natural language output that is closely related to the underlying data.

[0376] (5) In terms of processing production equipment operation data or transaction data, traditional systems mostly adopt offline analysis methods and do not design an automated pipeline at the system level that is "collected in chronological order - missing data completion - outlier deletion - indicator calculation - output in association with natural language description and visualization results". This makes it difficult for computers to provide users with data views and explanatory texts that can be directly used for decision-making in real-time or near-real-time scenarios.

[0377] (6) For application scenarios such as location-based security notifications, existing technologies lack a general mechanism that allows computers to simultaneously consider regional risk information extracted from datasets and user emotional states, automatically generate notification prompts with warning or reassurance attributes, and generate personalized natural language notification texts through generative artificial intelligence models. As a result, it is difficult to intelligently optimize information push strategies and content expression methods through computer systems.

[0378] In summary, how to build an integrated technical framework within a computer system that integrates multi-source data, calculates evaluation indicators, infers emotional states, automatically generates prompts, and invokes generative artificial intelligence models, enabling the server to efficiently, robustly, and scalably complete data processing, dynamic adjustment of indicators, and automatic generation of natural language documents without extensive human intervention, has become a pressing technical challenge in this field.

[0379] The specific processing performed by the specific processing unit 290 of the data processing apparatus 12 in Application Example 2 is achieved by the following means.

[0380] In this invention, the server includes: means for acquiring structured and unstructured data from multiple public and civilian information sources by an information processing device, and integrating the structured and unstructured data according to identifiers to generate an integrated data set; means for calculating evaluation indicators related to social infrastructure construction, funding provision, business evaluation, or production management from the integrated data set using statistical processing and machine learning methods, and summarizing the evaluation indicators by time or regional units; means for inferring emotional states based on user behavior information, voice information, image information, or text information acquired from sensing devices or input / output devices; and means for applying data transformation and visualization methods to the integrated data set and the evaluation indicators to generate tabular and graphical data, and for combining the tabular and graphical data... The following methods are employed: methods for automatically placing data into document templates; methods for dynamically adjusting target values ​​or evaluation benchmarks based on the evaluation indicators and the emotional state; methods for generating prompt statements for input into a generative artificial intelligence model based on the integrated data set, the evaluation indicators, the tabular data, the graphical data, and the emotional state, and using the prompt statements to enable the generative artificial intelligence model to perform natural language generation processing, thereby automatically generating application documents, report documents, or supplementary information; and methods for acquiring operational data or transaction data from production equipment or information terminals in chronological order, performing missing value completion, outlier deletion, and indicator calculation on the operational data or transaction data, and associating the natural language description generated by the generative artificial intelligence model with the visualization results of the operational data or transaction data and outputting it to the user terminal. This allows for the formation of a unified processing chain within the server, encompassing multi-source data integration, indicator calculation, sentiment inference, automatic construction of prompts, and invocation of generative artificial intelligence models. This enables automated analysis and document generation of large-scale heterogeneous data with minimal human intervention. The server can dynamically adjust evaluation objectives and output content complexity based on sentiment states, enhancing the adaptability and understandability of human-computer interaction. Furthermore, by tightly coupling time-series data processing, visualization results, and natural language explanations, the processing efficiency and technical performance of the computer system in data-driven decision-making scenarios are significantly improved.

[0381] "Public information sources" refer to data sources provided by public institutions, administrative agencies, or other public organizations that are publicly available or provided within authorized scope, including but not limited to storage systems or services for population statistics, financial information, infrastructure operation information, security information, and related statistical data.

[0382] "Civilian information sources" refer to non-public data sources operated by enterprises, public institutions, social organizations, or individuals for providing business data or user data, including but not limited to storage systems or services for commercial transaction data, consumer behavior data, production operation data, equipment log data, and market statistics data.

[0383] "Structured data" refers to data that has a predefined schema, can be organized in rows and columns and represented by fixed fields, including but not limited to records stored in a database in tabular form and datasets with explicit field names and values.

[0384] "Unstructured data" refers to data that does not conform to a predefined table structure or pattern, including but not limited to natural language text data, image data, audio data, video data, and other data information that is difficult to represent directly using table fields.

[0385] "Identifier" refers to tagging information used to associate the same entity or object between different data sources, including but not limited to unique number, code field, name field, timestamp field or a combination thereof.

[0386] An "integrated dataset" refers to a dataset obtained by associating and merging structured and unstructured data from multiple public and private information sources using identifiers or matching rules. This dataset can be used for unified analysis and processing.

[0387] "Statistical processing methods" refer to program modules or algorithms used to perform statistical operations on data, such as aggregation, grouping, summation, averaging, variance calculation, and correlation analysis.

[0388] "Machine learning methods" refer to program modules or algorithms that automatically build models based on sample data and perform prediction, classification, clustering or regression, including but not limited to supervised learning, unsupervised learning and reinforcement learning models.

[0389] "Evaluation indicators" refer to quantitative values ​​or classification results used to measure the degree of achievement of goals such as social infrastructure construction, funding provision, business evaluation, or production management, including but not limited to ratios, scores, grades, or comprehensive indices.

[0390] "Time unit" refers to the unit of division used to summarize or statistically analyze data on a timeline, including but not limited to hour, day, week, month, quarter, or year.

[0391] "Regional unit" refers to a geographical area or spatial range used to divide and summarize data in a spatial dimension, including but not limited to administrative regions, blocks, grid units, or geofences.

[0392] "Sensing device" refers to a hardware device used to collect physical quantities or environmental information, including but not limited to camera equipment, microphone equipment, position sensors, temperature sensors, pressure sensors and other detection equipment.

[0393] "Input / output device" refers to hardware or software components used to receive user input or present output information to the user, including but not limited to keyboards, touch screens, displays, speakers and corresponding user interface programs.

[0394] "User behavior information" refers to data that reflects a user's operating patterns and interaction records in the system, including but not limited to click records, browsing paths, input habits, dwell time, and operation sequences.

[0395] “Voice information” refers to the digital representation of sound signals emitted by a user or the environment, including but not limited to voice commands, natural conversational speech and its audio characteristics.

[0396] "Image information" refers to static or dynamic image data acquired through imaging devices, including but not limited to face images, scene images, surveillance video frames and their feature representations.

[0397] “Text information” refers to natural language data or symbolic data that exists in the form of character sequences, including but not limited to user input text, document content, message content, and log text.

[0398] "Emotional state" refers to the psychological tendency or emotional category of a user inferred based on user behavior information, voice information, image information, or text information, including but not limited to stress state, anxiety state, neutral state, and positive emotional state.

[0399] "Emotion inference methods" refer to program modules or models used to extract features from user behavior information, voice information, image information, or text information and infer emotional state, including but not limited to rule-based methods and machine learning-based methods.

[0400] "Data transformation methods" refer to program modules or algorithms that process raw data by performing format transformation, field mapping, normalization, encoding conversion, etc., in order to facilitate subsequent analysis and display.

[0401] "Visualization tools" refer to program modules or tools that convert numerical data, statistical results, or analytical results into charts, graphs, or other visual representations, including but not limited to components that generate line charts, bar charts, pie charts, and heat maps.

[0402] "Tabular data" refers to numerical or textual data that is organized in a row and column structure and can be displayed in tabular form, including but not limited to data tables and statistical result tables in reports.

[0403] "Graphical data" refers to visualization results represented in the form of graphics or images, including but not limited to static image files, vector graphics, interactive charts and their metadata.

[0404] A "document template" is a document framework with a predefined structure, layout, and placeholders, used to automatically populate data and text content during the generation process.

[0405] "Target value" refers to the expected value or threshold set in advance in the evaluation indicator dimension, which is used to compare and judge the actual data.

[0406] "Evaluation benchmark" refers to a set of standards, thresholds or rules used to judge the quality or completion of evaluation indicators, which may include a single value, range or logical condition.

[0407] "Indicator adjustment methods" refer to program modules or algorithms that dynamically modify the target value or evaluation benchmark of evaluation indicators based on emotional state or other conditions.

[0408] "Generative AI models" refer to AI models that can automatically generate text, images, or other content based on input information, especially models used for natural language generation.

[0409] "Prompt statements" refer to the instruction text input into a generative artificial intelligence model that instructs the model what type of content to generate or in what style to generate it.

[0410] "Natural Language Generation Processing" refers to the process by which generative artificial intelligence models automatically output natural language text based on prompts and input data.

[0411] "Application documents" refers to formal documents used to submit applications for funding, projects, or other review requests, including but not limited to application forms, instructions, and supporting materials.

[0412] "Report documents" refer to documents used to report on business progress, operational status, or evaluation results, including but not limited to periodic reports, summary reports, and analysis reports.

[0413] "Supplementary materials" refers to additional documents or materials used to support or explain the content of application documents and report documents, including but not limited to tables, figures, explanatory documents and data lists.

[0414] "Production equipment" refers to the equipment or system that performs operations during the production or manufacturing process, including but not limited to production line equipment, processing equipment and their control devices.

[0415] "Information terminal" refers to a computing device used to communicate with a server and display results, including but not limited to personal computers, mobile terminals and network terminal devices.

[0416] "Operational data" refers to data related to status, performance, or events generated during the operation of production equipment or information systems, including but not limited to equipment status logs, production data, and error records.

[0417] "Transaction data" refers to data related to transaction behavior generated in economic activities or business processes, including but not limited to order data, payment records, and settlement information.

[0418] "Missing information completion" refers to the process of inferring and filling in missing values ​​or fields in runtime or transaction data, including but not limited to interpolation based on statistical methods or models.

[0419] "Outlier removal" refers to the process of identifying and removing observations that significantly deviate from the normal range in order to improve the accuracy of data analysis.

[0420] "Natural Language Description" refers to text content that explains and describes data, charts, or analysis results in natural language.

[0421] "User terminal" refers to a terminal device that allows users to receive system output results and perform interactive operations, including but not limited to mobile terminals, desktop terminals, and web clients with display interfaces.

[0422] "Location information" refers to data information used to indicate the location of an object or user in space, including but not limited to longitude, latitude, altitude, and corresponding area identifiers.

[0423] "Crime occurrence information" refers to data such as the time, location, type, and frequency of illegal or irregular events, used to reflect the regional security situation.

[0424] "Safety-related information" refers to various types of data used to assess the safety of personnel or assets, including but not limited to security conditions, distribution of surveillance facilities, and emergency resource information.

[0425] "Infrastructure operation information" refers to various types of data used to reflect the operating status and performance of infrastructure systems, including but not limited to load, fault, and maintenance records of transportation facilities, energy facilities, or communication facilities.

[0426] "Warning messages" refer to messages used to remind users of potential risks or adverse situations, including but not limited to risk warnings, precautions, and alerts.

[0427] "Reassurance messages" refer to messages designed to alleviate users' anxiety and enhance their sense of security or confidence, including but not limited to positive statements, risk mitigation measures, and descriptions of trend improvement.

[0428] "Notification prompts" refer to prompts specifically designed for generating natural language text for notifications. Their content includes at least the subject of the notification, the required tone, and instructions on whether to include warning or reassuring information.

[0429] "Natural Language Notification Text" refers to the notification message content expressed in natural language, generated by a generative artificial intelligence model based on notification prompts and related data.

[0430] "Geographic information display" refers to the display method of presenting location information and related data in the form of a map or spatial layout on a graphical interface, including but not limited to electronic maps, regional color maps, and location marker maps.

[0431] The embodiments of the present invention will be described based on the information processing system structure described in Appendix 1 to 3. In the following description, the subjects are limited to "server," "endpoint," and "user," and the technical descriptions will focus on specific hardware, software, data structures, algorithm processes, and the internal processing of generative artificial intelligence models.

[0432] I. System Overall Structure and Hardware Configuration A server comprises one or more computing devices, each including at least a central processing unit (CPU), an optional graphics processing unit (GPU), main memory, non-volatile memory, and a network interface. The server connects to multiple endpoints and external data sources (public and private information sources) via wired or wireless networks.

[0433] A terminal includes electronic devices with display and input devices, such as mobile terminals, tablet terminals, or desktop terminals. Terminals typically include camera devices, microphone devices, positioning devices (such as satellite positioning modules), touch screens, etc., used to collect user behavior information, voice information, image information, and location information.

[0434] Users interact with the server through their devices to trigger operations such as data querying, report generation, and notification display.

[0435] II. Server-side program and module composition Servers run various software components on an operating system (such as a general-purpose operating system), including: - Data Acquisition and Integration Module: Uses a network library to acquire data from public and private information sources; - Data storage module: Runs a relational database management system, such as a general-purpose relational database; - Data processing module: Runs an interpreted language environment (e.g., Python), and is equipped with data processing libraries (e.g., Pandas), numerical computing libraries (e.g., NumPy), and statistical libraries (e.g., SciPy); - Machine Learning and Sentiment Inference Module: Uses machine learning libraries (such as Scikit-learn) and deep learning frameworks (such as TensorFlow or similar frameworks). - Visualization module: Uses graphics plotting libraries (such as Matplotlib) and visualization tool interfaces (such as the General Spreadsheet API or the Visualization Platform API). - Document generation module: Uses document processing libraries (such as libraries that can generate document formats) and layout engines (such as PDF generators). - Generative AI Model Interface Module: Enables calling generative AI models via network API or deploying the corresponding model locally; - Notification and Interface Module: Responsible for pushing notification messages to end devices and generating results.

[0436] The server uses these modules to form internal data flows and functional module links, making the various "means" described in the notes technically executable.

[0437] III. Implementation Forms of Multi-Source Data Acquisition and Integration When the server retrieves structured and unstructured data from public and private information sources, it uses a unified abstract data structure. The server converts data from open interfaces in formats such as JSON, CSV, and XML into an internal standard structure. - For structured data, the server maps it to multiple tables in a relational database, such as "Public Statistics Table", "Business Statistics Table", "Infrastructure Operation Table", "Security Incident Table", etc. - For unstructured data, the server stores text as text fields or full-text indexes, and stores images and audio as binary objects, while recording corresponding metadata (time, source, region identifier, etc.).

[0438] The server uses Pandas to load the above table into a DataFrame, defining a uniform identifier field (e.g., region code, facility code, user code) as the primary key within the DataFrame. The server then uses DataFrame merging operations (e.g., joining by identifier) ​​to form an integrated data set. Because the same identifier scheme is used uniformly in the database layer and in-memory data structures, the server can efficiently retrieve cross-source data in a single query, reducing the communication overhead of multiple remote accesses. This reduces network load and I / O overhead at the system level, thereby improving data management.

[0439] IV. Evaluation Indicator Calculation and Technical Effectiveness When calculating evaluation metrics, the server not only performs simple summarization, but also uses statistical processing and machine learning methods to perform pattern recognition and prediction on the integrated data set.

[0440] In terms of statistical processing, the server uses aggregation operations (summation, mean, median), variance and standard deviation calculations, and correlation coefficient calculations to generate basic indicators at the granularity of regional or time units. For example: - The server aggregates population, budget, and consumption into "budget per capita" and "consumption per capita"; - The server combines the number of security events with the density of camera devices to calculate the "event rate" and "camera coverage" for each area; The server resamples the production unit's operating data by time series to calculate hourly and daily output, defect rate, and downtime rate.

[0441] In machine learning, servers can employ various model architectures, such as: - The server uses linear regression or ridge regression models to model features such as "infrastructure investment", "population change" and "consumer behavior" to predict "future consumption" or "future service coverage". The server uses a decision tree model or a random forest model to classify regional security risks into low, medium, and high categories, using features such as "historical crime rate," "camera coverage," and "nighttime pedestrian traffic."

[0442] The server trains the model offline using the training dataset, employing mean squared error or cross-entropy as the loss function and updating the model weights using gradient descent or its variants. During inference, the server loads the trained model and quickly calculates evaluation metrics or risk levels for new data. Because it uses model-based predictions rather than simple rules, the server can make more accurate judgments about future trends and implicit risks, thereby improving the overall system's prediction quality and decision support capabilities with the same hardware resources.

[0443] V. Emotion Inference Methods and Feature Engineering The server uses a deep neural network model for sentiment inference.

[0444] The end-to-end system converts the acquired speech signal into Mel spectrum or MFCC features and converts the image information into a standard resolution face image.

[0445] After receiving these features, the server inputs them into the sentiment classification model.

[0446] In one embodiment, the server uses a convolutional neural network (CNN) combined with a bidirectional recurrent neural network (Bi-LSTM) structure: - For speech, the server first extracts local temporal features through a one-dimensional convolutional layer, then captures long-range dependencies through a bidirectional LSTM, and finally outputs the probability distribution of several emotion categories (such as "calm", "stress", "anxiety" and "pleasure") through a fully connected layer. For images, the server uses two-dimensional convolutional layers to extract local texture and facial expression features, downsamples them through pooling layers, and then outputs the emotion category probability through multiple fully connected layers.

[0447] During training, the server uses labeled audio and video datasets, employs cross-entropy as the loss function, updates model parameters through backpropagation, and can utilize data augmentation techniques (such as random cropping, rotation, and noise addition) to enhance robustness. Compared to traditional rule-based sentiment assessment (based on simple thresholds and manual rules), this deep learning structure can more accurately determine the sentiment state under complex and noisy conditions, thus providing precise sentiment input for subsequent metric adjustments and prompt generation.

[0448] VI. Dynamic Adjustment of Indicators and Non-Standard Rules When adjusting metrics, the server does not simply make linear proportional adjustments, but rather employs a non-traditional dynamic correction strategy based on sentiment levels. For example: - When the server determines that a user is in a "positive emotional state" and has stable historical performance, the server can increase the target value of several evaluation indicators by a certain percentage (e.g., +10%), while keeping other key safety indicators unchanged. When the server determines that a user is in a "stressed" or "anxious" state, it not only lowers the target values ​​of certain metrics, but also automatically selects a simpler display dimension, that is, hides temporarily irrelevant detailed fields and only retains the core metrics.

[0449] The server uses a pre-defined rule graph (such as a two-dimensional mapping table based on emotion categories and indicator types) and optional reinforcement learning algorithms to determine how to combine and adjust multiple indicators in different emotional and business environments. This approach, which incorporates emotional states into the dynamic design of indicators, differs from traditional systems with fixed target values. It is an adaptive indicator setting technology that combines emotion perception, which can significantly reduce the cognitive load on users under high-pressure situations and improve interaction efficiency.

[0450] VII. Technical Application of Prompt Statement Generation and Generative Artificial Intelligence Models Before calling the generative artificial intelligence model, the server encodes the integrated dataset, evaluation metrics, tabular data, graphical data summaries, and sentiment states into a structured intermediate representation, such as an internal object containing several slots (e.g., "scene type", "key values", "trend description", "emotional label", "output tone", etc.).

[0451] Based on this intermediate representation, the server generates a prompt statement using rule templates and a semantic concatenation algorithm. In one implementation, the server employs the following strategy for constructing the prompt statement: - The server selects different "role settings" based on the scenario (e.g., "production management consultant", "public policy expert", "personal financial advisor", "security reminder assistant"). - The server inserts key data into specific statement fragments (e.g., "Total output is X pieces, average defect rate is Y%)"); The server maps emotion tags to tone control words (such as "gentle tone", "moderate appeasement", "formal executive style").

[0452] For example, the prompt statement constructed by the server in a production line scenario could be: "You are a production management consultant. Below are the statistics for production line A over a certain period of time:" 1) Total output: 300,000 units; 2) Average defect rate: 1.3%; 3) Maximum daily output: 12,000 pieces; 4) Minimum daily output: 8,500 pieces.

[0453] Based on this data, please generate a summary for factory management, outlining the overall performance of the production line, key issues, and three improvement suggestions. The summary should be no more than 800 words. In the context of funding application, the prompt message constructed by the server could be: "You are a public policy expert. Below is key data from a certain autonomous region:" - Population: 100,000; - Annual budget: 10 billion; - Current annual total consumption: 5 billion; - Target KPI: Increase annual consumption by 10% within the next 3 years.

[0454] Please write a 'Project Background and Necessity Explanation' section for the application for funding. The section should be in a formal administrative document style and include the following structure: 1) Current Situation Analysis; 2) Existing Problems; 3) Necessity of Funding Support. Approximately 1500 words. In the context of personal spending, the prompt message constructed by the server could be: "You are a personal financial advisor. Below is a user's spending history over the past three months:" - Average total monthly expenditure: 20,000 yuan; - Total expenditure this month: 26,000 yuan (30% higher than the average); - Food and beverage expenses this month: 9,000 yuan (20% increase compared to last month); - Entertainment expenses this month: 5,500 yuan (an increase of 35% compared to last month).

[0455] User sentiment analysis result: Anxiety.

[0456] Please generate an expenditure analysis and savings suggestion report for this user. Requirements: 1) First, explain the current expenditure situation in plain language; 2) Identify the main categories of abnormal expenditures; 3) Provide 3-5 specific and actionable suggestions for saving money; 4) Use a gentle tone and be mindful of calming the user. Because the prompts are automatically generated internally on the server based on a structured intermediate representation, this approach avoids the unpredictability of manually written prompts, making the output of generative AI models more predictable and repeatable in terms of content structure, style, and data consistency. This constitutes a technical calling layer for generative AI models, representing an improvement to the computer natural language generation process.

[0457] VIII. Time Series Data Processing and Visualization Integration The server stores operational and transaction data collected from production facilities or information terminals using a timestamp-indexed data structure. During data processing: - Perform interpolation or fill in missing time points with the most recent value; - Identify, label, or remove outliers using statistical methods (such as based on standard deviation range or box plot thresholds); - Perform sliding window calculations on the cleaned time series to obtain smoothing indices such as moving average and moving variance.

[0458] The server directly uses these metrics as parameters or explanatory text input to the generative AI model and simultaneously displays them graphically. The server establishes an explicit association between the generated natural language explanatory text and the corresponding graphical data (e.g., through internal IDs or metadata links), ensuring that "each explanatory text corresponds to one or more specific charts" when displayed on the client. Users can then highlight the corresponding chart when they click on an explanatory text on the client. This interconnected structure requires the server to maintain synchronization in data structure design and front-end / back-end interfaces, achieving integrated binding between the data view and the natural language interpretation. This technical architecture is significantly superior to the traditional approach of "independently generating charts and independently writing explanatory texts," reducing mismatches and maintenance costs between different information sources.

[0459] IX. Location-Based and Emotion-Based Security Notification Implementation Models After acquiring the user's location information, the client sends it to the server. The server maps the latitude and longitude to regional unit codes and reads crime occurrence information, security-related information, and infrastructure operation information for that region from the integrated dataset. The server then reduces this information to form a regional security summary (including the number of incidents, trends, facility coverage, etc.).

[0460] The server combines emotional states to generate notification prompts, for example: "You are a safety reminder assistant for ordinary citizens."

[0461] Regional Information: - Area Name: XX Street; - Recent crime statistics: There were 15 nighttime burglaries in the past three months, a 20% decrease compared to the same period last year; - Overall safety score: 45 (slightly below the city average of 60).

[0462] User sentiment: Anxious.

[0463] Please generate a notification of approximately 300 words, with the following requirements: 1) Remind users to be careful at night and avoid walking alone on secluded paths; 2) It also states that the police have increased patrols and the total number of cases is decreasing, in order to alleviate users' anxiety; 3) The language is concise and user-friendly. The server connects the generated natural language notification text to the map display module, highlighting the current area and recommended safe routes on the map. Unlike traditional static announcements, this notification content is tightly coupled with real-time data and emotional information at both the text and spatial display levels, thereby enhancing the targeted nature of the warning and its reassuring effect.

[0464] 10. Technical Effects and Improvements in Computer Technology The server achieves the following technical effects through the aforementioned techniques, including structured data integration, statistical and machine learning modeling, sentiment inference, automatic generation of prompts, and visual binding: - At the data management level, by using unified identifiers and intermediate data structures, we can reduce multiple redundant queries and format conversions, reduce storage and communication overhead, and improve the speed of data reading and processing. - At the computational level, through specially designed time series processing, anomaly detection, and machine learning inference processes, the accuracy and stability of indicator calculations are improved, and the errors caused by manually set rules are reduced. - At the model invocation level, by automatically constructing prompt statements and combining them with structured intermediate representations, the output of generative artificial intelligence models becomes more controllable and repeatable, avoiding the unpredictability and instability of traditional manual prompt statements, thereby improving the technical quality of natural language generation. - At the interaction level, through emotion inference and indicator adjustment mechanisms, the server changes the complexity of the output content and the target value settings according to the user's emotional state. This is not a simple automation of human tasks, but a technological improvement that dynamically adjusts the calculation procedures and output strategies inside the computer to achieve human-computer interface adaptation and cognitive load control. - In terms of overall system performance, the server leverages a unified data flow and modular processing architecture to enable multi-source data analysis, model inference, and document generation to be completed in the same chain, reducing redundant data loading and multi-process communication, and improving throughput and response speed.

[0465] XI. Alternative Implementation Methods and Extensions The server can replace some technical components in different implementation forms: - Servers can use different types of databases (relational or time-series databases) to store integrated data sets; - Servers can implement sentiment inference or risk rating using different deep learning architectures (such as Transformer architecture or graph neural networks), as long as these models have well-defined input features, loss functions, and training methods; - The server can deploy generative AI models on a local computing cluster or access them through cloud services, as long as the prompt statement construction and the call flow remain consistent. The server can extend evaluation metrics and prompt templates for different application areas (such as medical data analysis, energy system monitoring, etc.) without changing the underlying data integration and processing framework.

[0466] The terminal can also be a vehicle-mounted terminal, a wearable terminal, etc., as long as it can collect the necessary sensor data and display the server output results.

[0467] Through the above-mentioned various implementation forms, the system of the present invention can not only directly link with production equipment, safety facilities and user terminals in real-world scenarios, but also realize an integrated technical framework of data flow, model reasoning and document generation within the computer, thereby improving existing computer technology in terms of processing speed, accuracy, data management and controllability of natural language generation.

[0468] use Figure 14 The processing flow is explained.

[0469] Step 1: The server obtains raw data from multiple public information sources and multiple civilian information sources.

[0470] The input is a raw data stream of structured data (such as tables and records) and unstructured data (such as text, images, and audio) from an external interface.

[0471] The server sends requests to external data services through a network interface, receives data files in JSON, CSV, or XML format, and reads text files, image files, and audio files from the file system or object storage.

[0472] The server identifies the format of the received data, parses structured data into field records, and saves unstructured data as binary objects and their metadata.

[0473] The output consists of multiple raw data tables stored in the database and their corresponding unstructured data storage entries, with each record containing a source identifier and a timestamp.

[0474] Step 2: The server performs unified identifier mapping and data cleaning on the data obtained in step 1.

[0475] The input consists of multiple raw data tables and related metadata, including different field names and encoding methods.

[0476] According to the preset mapping rules, the server converts fields such as region name, device number, and user number in each data table into unified standard identifiers, such as standard region code or standard device code; the server uses a data processing library to fill in missing fields, correct or delete records with incorrect formats, and mark or remove significant outliers.

[0477] The server adds a unified "primary key column" to the data frame for cross-table joins.

[0478] The output consists of a set of cleaned and uniformly encoded intermediate data tables, and corresponding data frames. Each record contains a uniform identifier and cleaned field values.

[0479] Step 3: The server integrates the cleaned multi-source data to generate an integrated data set.

[0480] The input consists of multiple intermediate data tables or data frames obtained in step 2.

[0481] The server performs table join operations based on a unified identifier, merging records from different public and private information sources into a unified data structure. For example, it can merge population data, budget data, security incident data, and consumption data by regional code, or merge production equipment operation data and maintenance records by equipment code.

[0482] The server fills in missing combinations and stores one-to-many relationships in aggregate or list form to ensure that each integrated record fully reflects the multidimensional characteristics of a certain entity or region.

[0483] The output is a data table (or data frame) of one or more integrated data sets, each containing multidimensional feature fields that describe the same entity or region.

[0484] Step 4: The server uses statistical processing and machine learning techniques to calculate evaluation metrics.

[0485] The input is the integrated data set generated in step 3.

[0486] The server groups and aggregates data by time unit (e.g., day, month, year) and regional unit (e.g., administrative region, factory production line), and calculates basic statistics (e.g., total, average, proportion, rate of change). The server uses these aggregation results as features to input into a trained machine learning model, such as a regression model or a classification model, to predict future values ​​(e.g., future consumption, future output) or risk levels (e.g., regional security level).

[0487] The server uses a numerical computation library to perform matrix operations and vectorization during the computation process, thereby improving computation speed.

[0488] The output is a table of evaluation indicators for each time unit or regional unit, including fields such as current indicator value, predicted value, and risk level.

[0489] Step 5: The device collects user behavior information, voice information, and image information from each device and sends them to the server.

[0490] The input consists of user operation records on the device, voice input, and image or video frames captured by the camera device.

[0491] The device records user click sequences, dwell time, input text and other behavioral data, uses a microphone to collect audio data, and uses a camera to collect facial or upper body images. If necessary, it performs preliminary feature extraction on the audio and images (such as calculating audio spectrograms and cropping facial regions).

[0492] The endpoint sends the aforementioned behavioral data and raw or preprocessed feature data to the server through a secure communication channel.

[0493] The output is a set of data packets containing behavioral sequences, audio clips, and image frames (or features thereof) sent to the server.

[0494] Step 6: The server infers emotional state based on user behavior information, voice information, and image information.

[0495] The input consists of the behavioral data, audio data, and image data uploaded by the client in step 5.

[0496] The server extracts acoustic features (such as MFCC, energy, and fundamental frequency) from audio data, extracts facial features from image data (through the output of the intermediate layer of a convolutional neural network), and concatenates behavioral features with audio and video features into a multimodal feature vector.

[0497] The server inputs the feature vector into a pre-trained deep neural network model (such as an emotion classification network composed of convolutional layers, recurrent layers, and fully connected layers), calculates the probability distribution of each emotion category, and determines the user's current emotional state based on the maximum probability or threshold rule, such as "calm," "stressed," "anxious," or "satisfied."

[0498] The output consists of sentiment state labels and their confidence scores, which are stored in a sentiment state record table for subsequent indicator adjustments and prompt statement generation.

[0499] Step 7: The server summarizes and standardizes the evaluation indicators by time unit and region unit.

[0500] The input consists of the evaluation index table obtained in step 4 and the emotional state information obtained in step 6 (optional).

[0501] The server selects the set of indicators to be displayed or used for decision-making according to business rules, and performs normalization or standardization processing on the indicators (e.g., mapping the indicators to the 0-1 range or Z score) so that they can be compared between different units of measurement; the server summarizes the indicators by time scales such as daily, weekly, and monthly, and calculates the trend change value (such as the difference between the current period and the previous period or the growth rate).

[0502] When it is necessary to make horizontal comparisons across different regions or devices, the server sorts and calculates quantiles for each indicator to generate a unified score.

[0503] The output is a set of standardized evaluation indicators with multiple scales and dimensions, including raw values, standardized values ​​and ranking information.

[0504] Step 8: The server dynamically adjusts the target value and evaluation benchmark based on the emotional state.

[0505] The input consists of the standardized evaluation index set generated in step 7 and the emotional state labels from step 6.

[0506] The server maintains an internal sentiment-policy mapping rule table to map different sentiment states to different adjustment policies (such as raising the target value, lowering the target value, or keeping it unchanged).

[0507] When the emotional state is positive, the server increases the target value of some upgradable indicators (such as output increase and service coverage increase) according to a predetermined ratio; when the emotional state is stress or anxiety, the server lowers the target value or widens the qualified range, and marks some secondary indicators as "hidden" to reduce the amount of output information.

[0508] The server writes the adjusted target values ​​and evaluation benchmarks into the indicator configuration table for subsequent display and document generation.

[0509] The output includes the updated target value table and evaluation benchmark table, as well as the corresponding policy application log.

[0510] Step 9: The server applies data transformation and visualization processing to the integrated datasets and evaluation metrics, generating tabular and graphical data.

[0511] The input consists of the integrated data set from step 3 and the evaluation indicators and target values ​​from steps 7 and 8.

[0512] The server uses a data transformation module to convert internal data frames into a table structure suitable for documents and front-end display, such as a two-dimensional array or a list of key-value pairs; the server selects the appropriate chart format for different indicators based on the chart type selector (such as a line chart for time trends, a bar chart for comparison, and a pie chart for composition ratios).

[0513] The server uses a visualization library to generate corresponding graphical data files (such as PNG images or vector graphics) and records the associated indicator ID and time / region range for each chart, so that subsequent text descriptions can accurately correspond to the charts.

[0514] The output consists of a table data structure for document population and a set of graphic data files, as well as metadata relating the tables and graphics.

[0515] Step 10: The server generates prompts for generative artificial intelligence models.

[0516] The input consists of summary information of the integrated dataset, evaluation indicators and their target values, tabular data, metadata of graphical data, and sentiment state labels.

[0517] The server selects the corresponding set of prompt statement templates based on the scenario type (e.g., production management, fund application, personal expenditure analysis, safety notification); the server maps key values ​​(e.g., total output, defect rate, budget, KPI, regional safety score) and emotional state into natural language fragments, and splices them into complete prompt statements in a predetermined order.

[0518] The server explicitly specifies the role setting, output style, structure requirements, and word count constraints in the prompt statement. For example, it may require the use of a formal administrative style or a friendly and reassuring tone, and require the output to include several suggestions or conclusions.

[0519] The output is a text message or prompt, each of which is associated with a specific data entity (such as a production line, a region, or a user).

[0520] Step 11: The server invokes a generative artificial intelligence model to perform natural language generation processing.

[0521] The input consists of the prompt statement generated in step 10 and optional supplementary data snippets (such as raw indicator values ​​or data summaries).

[0522] The server sends the prompts to the generative artificial intelligence model through the interface module. This model can be a pre-trained language model based on the Transformer architecture. When the server is invoked, it sets model parameters, such as the upper limit of the generation length, temperature parameters, and sampling strategies, to control the diversity and stability of the output.

[0523] Generative AI models parse instructions and embedded data based on prompts to generate natural language text containing background information, data interpretation, trend analysis, and suggestions.

[0524] The server receives the returned text, performs simple format validation (such as removing irrelevant characters and checking for obvious errors), and binds it to the corresponding data entity and chart metadata.

[0525] The output is a set of natural language explanatory text that can be directly used for document or interface display.

[0526] Step 12: The server populates tabular data, graphical data, and natural language descriptive text into document templates to generate application documents, report documents, or supplementary materials.

[0527] The input consists of the tabular and graphical data generated in step 9, the natural language text generated in step 11, and a predefined document template.

[0528] The server scans the placeholders in the document template, inserts table data into the table area at the specified position, inserts graphic data into the chart area, and inserts natural language text into the introduction, analysis, or recommendations sections; the server automatically adjusts paragraphs, headings, and pagination according to the template's layout rules.

[0529] The server exports the completed document in a distributable format (such as PDF or editable document format) and saves the file in the storage system, while also recording the file path and index information.

[0530] The output is a complete document file that users can download or view online, along with index records associated with the document.

[0531] Step 13: The server generates a visual explanation interface based on runtime or transaction data and sends the results to the client.

[0532] The input consists of the processing results of time series running data or transaction data, the corresponding chart files, and the associated natural language explanatory text.

[0533] The server builds the front-end display configuration, binds each natural language description to one or more chart objects, and sends response data containing chart URLs, description text, and interactive configurations to the client through an interface.

[0534] The server can specify interactive behaviors in the configuration, such as highlighting the corresponding chart or highlighting specific data points on the chart when a user clicks on a piece of text on the client.

[0535] The output consists of a set of interface configuration data sent to the client and corresponding visual resource links.

[0536] Step 14: The client displays reports or notifications based on the data provided by the server.

[0537] The input consists of the interface configuration data, chart links, and natural language text sent by the server in step 13.

[0538] The app renders tables and charts on the screen and displays natural language text in a modular layout. It also implements interactive functions such as clicking, highlighting, and switching views, allowing users to view corresponding charts and data details while reading instructions.

[0539] When the client receives a security notification, it displays the current location information and security level on the map interface, and shows the generated natural language notification text on the interface.

[0540] The output is a unified graphic and text interface presented to the user, as well as further user actions on the interface.

[0541] Step 15: Users can view and use the results generated by the server on their devices.

[0542] The input consists of report document snippets, charts, and notification information displayed on the client.

[0543] Users can understand indicator changes and system analysis conclusions by reading natural language descriptions and observing charts, and then decide whether to adjust production plans, modify funding application schemes, change personal spending habits, or change travel routes.

[0544] Users can input feedback or request new analysis on the client side. The client side sends this feedback as new behavioral information to the server for subsequent iterative analysis and sentiment inference.

[0545] The output consists of user operation feedback and behavior records. This data will be used as new inputs in step 5 and subsequent steps, forming a closed-loop processing flow.

[0546] The specific processing unit 290 sends the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires sound representing user input regarding the result of the specific processing. The control unit 46A sends the sound data representing user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the sound data.

[0547] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 are prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers from the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms, such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, 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, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0548] Furthermore, the processing of the aforementioned data processing system 10 is performed 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 performed by both the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, 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.

[0549] For example, the collection unit is implemented by the control unit 46A of the smart device 14 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the smart device 14 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the output device 40 of the smart device 14 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0550] In the above embodiments, examples of specific processing 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.

[0551] Second Implementation Method Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0552] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. A server can be cited as an example of the data processing device 12.

[0553] 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. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0554] 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.

[0555] Microphone 238 receives instructions from user 20 by receiving sounds emitted by user 20. Microphone 238 captures sounds emitted by user 20 and converts the captured sounds into sound data, which is then output to processor 46. Speaker 240 outputs sound according to instructions from processor 46.

[0556] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and imaging elements such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, which captures images of the user 20's surroundings (e.g., the field of view defined by an angle equivalent to the field of vision of an average healthy person).

[0557] Communication I / F44 is connected to network 54. Communication I / F44 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 / F44 and 26 is performed in a secure state.

[0558] Figure 4 This illustrates an example of the main functions of the data processing device 12 and the smart glasses 214. For example... Figure 4 As shown, in the data processing device 12, specific processing is performed by the processor 28. The specific processing program 56 is stored in the memory 32.

[0559] The specific processing program 56 is an example of a "program" involved in the technology of 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 according to the specific processing program 56 executed on the RAM 30.

[0560] 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. In the emotion inference function (emotion-specific function) using the emotion-specific model 59, various inferences and predictions related to the user's emotions are performed, including inferences and predictions of the user's emotions, but this is not limited to this example. Furthermore, emotion inference and prediction may also include, for example, emotion analysis (parsing).

[0561] In the smart glasses 214, the processor 46 performs reception and output processing. The memory 50 stores the reception and output program 60. The processor 46 reads the reception and output program 60 from the memory 50 and executes the read reception and output program 60 on the RAM 48. The reception and output processing is implemented by the processor 46 operating as a control unit 46A according to the reception and output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 has the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and these models can also be used to perform the same processing as the specific processing unit 290.

[0562] Next, the specific processing of 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".

[0563] Example 1 The process is the same as that of the specific process described in Embodiment 1 in the first embodiment above, so the description is omitted.

[0564] Application Example 1 The process is the same as that in the specific processing described in Application Example 1 of the first embodiment above, so the description is omitted.

[0565] Example 2 The process is the same as that of the specific process in Embodiment 2 described in the first embodiment above, so the description is omitted.

[0566] Application Example 2 The process is the same as that in the specific processing described in Application Example 2 of the first embodiment above, so the description is omitted.

[0567] The specific processing unit 290 sends the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A outputs the result of the specific processing to the speaker 240. The microphone 238 acquires sound input representing the user's input regarding the result of the specific processing. The control unit 46A sends the sound 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 sound data.

[0568] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 are prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers from the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms, such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, 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, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0569] Furthermore, the processing of the aforementioned data processing system 10 is performed 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 performed by both 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.

[0570] For example, the collection unit is implemented by the control unit 46A of the smart glasses 214 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the smart glasses 214 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the speaker 240 of the smart glasses 214 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0571] In the above embodiments, examples of specific processing 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.

[0572] Third Implementation Method Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0573] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. A server can be cited as an example of the data processing device 12.

[0574] 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. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0575] 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.

[0576] Microphone 238 receives instructions from user 20 by receiving sounds emitted by user 20. Microphone 238 captures sounds emitted by user 20 and converts the captured sounds into sound data, which is then output to processor 46. Speaker 240 outputs sound according to instructions from processor 46.

[0577] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and imaging elements such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, which captures images of the user 20's surroundings (e.g., the field of view defined by an angle equivalent to the field of vision of an average healthy person).

[0578] Communication I / F44 is connected to network 54. Communication I / F44 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 / F44 and 26 is performed in a secure state.

[0579] Figure 6 This illustrates an example of the main functions of the data processing device 12 and the head-mounted terminal 314. For example... Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The specific processing program 56 is stored in the memory 32.

[0580] The specific processing program 56 is an example of a "program" involved in the technology of 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 according to the specific processing program 56 executed on the RAM 30.

[0581] The memory 32 stores the data generation model 58 and the emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0582] In the head-mounted terminal 314, the processor 46 performs the acceptance / output processing. The memory 50 stores the acceptance / output program 60. The processor 46 reads the acceptance / output program 60 from the memory 50 and executes the read acceptance / output program 60 on the RAM 48. The acceptance / output processing is implemented by the processor 46 operating as a control unit 46A according to the acceptance / output program 60 executed on the RAM 48.

[0583] Next, the specific processing of 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".

[0584] Example 1 The process is the same as that of the specific process described in Embodiment 1 in the first embodiment above, so the description is omitted.

[0585] Application Example 1 The process is the same as that in the specific processing described in Application Example 1 of the first embodiment above, so the description is omitted.

[0586] Example 2 The process is the same as that of the specific process in Embodiment 2 described in the first embodiment above, so the description is omitted.

[0587] Application Example 2 The process is the same as that in the specific processing described in Application Example 2 of the first embodiment above, so the description is omitted.

[0588] The specific processing unit 290 sends the result of the specific processing to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A outputs the result of the specific processing to the speaker 240 and the display 343. The microphone 238 acquires sound input representing the user's input regarding the result of the specific processing. The control unit 46A sends the sound 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 sound data.

[0589] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 includes prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, 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, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0590] Furthermore, the processing of the aforementioned data processing system 10 is performed 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 performed by both 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.

[0591] For example, the collection unit is implemented by the control unit 46A of the head-mounted terminal 314 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the head-mounted terminal 314 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12 to analyze the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12 to generate a menu using a generation AI. For example, the serving unit is implemented by the speaker 240 and display 343 of the head-mounted terminal 314 or the specific processing unit 290 of the data processing device 12 to provide the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0592] In the above embodiments, examples of specific processing 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.

[0593] Fourth Implementation Method Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0594] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. A server can be cited as an example of the data processing device 12.

[0595] 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. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0596] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a controlled object 443. 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, controlled object 443, and communication I / F 44 are also connected to the bus 52.

[0597] Microphone 238 receives instructions from user 20 by receiving sounds emitted by user 20. Microphone 238 captures sounds emitted by user 20 and converts the captured sounds into sound data, which is then output to processor 46. Speaker 240 outputs sound according to instructions from processor 46.

[0598] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and imaging elements such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, to photograph the area around robot 414 (e.g., the field of view defined by a perspective equivalent to the field of vision of an average healthy person).

[0599] Communication I / F44 is connected to network 54. Communication I / F44 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 / F44 and 26 is performed in a secure state.

[0600] The controlled object 443 includes a display device, LEDs (light-emitting diodes) for the eyes, and motors for driving the arms, hands, and feet. The posture or movement of the robot 414 is controlled by controlling the motors in the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, the robot 414's facial expressions can also be expressed by controlling the illumination state of the LEDs in its eyes.

[0601] Figure 8 This illustrates an example of the main functions of the data processing device 12 and the robot 414. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The specific processing program 56 is stored in the memory 32.

[0602] The specific processing program 56 is an example of a "program" involved in the technology of 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 according to the specific processing program 56 executed on the RAM 30.

[0603] The memory 32 stores the data generation model 58 and the emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0604] In robot 414, the processor 46 performs the acceptance and output processing. The memory 50 stores the acceptance and output program 60. The processor 46 reads the acceptance and output program 60 from the memory 50 and executes the read acceptance and output program 60 on RAM 48. The acceptance and output processing is implemented by the processor 46 acting as the control unit 46A according to the acceptance and output program 60 executed on RAM 48.

[0605] Next, the specific processing of 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".

[0606] Example 1 The process is the same as that of the specific process described in Embodiment 1 in the first embodiment above, so the description is omitted.

[0607] Application Example 1 The process is the same as that in the specific processing described in Application Example 1 of the first embodiment above, so the description is omitted.

[0608] Example 2 The process is the same as that of the specific process in Embodiment 2 described in the first embodiment above, so the description is omitted.

[0609] Application Example 2 The process is the same as that in the specific processing described in Application Example 2 of the first embodiment above, so the description is omitted.

[0610] The specific processing unit 290 sends the result of the specific processing to the robot 414. In the robot 414, the control unit 46A outputs the result of the specific processing to the speaker 240 and the controlled object 443. The microphone 238 acquires sound input representing the result of the specific processing. The control unit 46A sends the sound 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 sound data.

[0611] Data generation model 58 is a so-called generative AI (Artificial Intelligence). Examples of data generation models 58 include ChatGPT (registered trademark) (accessible via the internet (URL: https: / / openai.com / blog / chatgpt)). Data generation model 58 is obtained through deep learning on a neural network. Input to data generation model 58 are prompt words containing instructions, and inference data such as sound data representing sound, text data representing text, and image data representing images (e.g., still image data or animation data). Data generation model 58 infers from the input inference data based on the instructions represented by the prompt words and outputs the inference result in one or more data forms, such as sound data, text data, and image data. Data generation model 58 includes, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or induction. The specific processing unit 290 performs the aforementioned specific processing while using data generation model 58. The data generation model 58 can also be a model finely tuned to output inference results from prompts that do not contain instructions. In this case, the data generation model 58 can output inference results based on prompts that do not contain instructions. The data processing apparatus 12, etc., includes various data generation models 58, including AI other than the generation AI. AI other than the generation AI can be, for example, 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, and can perform various processes, but is not limited to this example. Furthermore, the AI ​​can also be an AI agent. Furthermore, when the processing of the above-mentioned parts is performed by AI, the processing can be performed partially or entirely by AI, but is not limited to this example. Furthermore, the processing performed by the AI ​​including the generation AI can be replaced by processing in the rule base, and the processing in the rule base can also be replaced by processing performed by the AI ​​including the generation AI.

[0612] Furthermore, the processing of the aforementioned data processing system 10 is performed 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 performed by both 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.

[0613] For example, the collection unit is implemented by the control unit 46A of the robot 414 or the specific processing unit 290 of the data processing device 12. For example, the acquisition unit uses the camera 42 or communication I / F 44 of the robot 414 to acquire step data, which is then processed by the specific processing unit 290 of the data processing device 12. For example, the analysis unit is implemented by the specific processing unit 290 of the data processing device 12, which analyzes the data from the collection unit and the acquisition unit. For example, the generation unit is implemented by the specific processing unit 290 of the data processing device 12, which uses a generation AI to generate a menu. For example, the serving unit is implemented by the speaker 240 of the robot 414 and the control object 443 or the specific processing unit 290 of the data processing device 12, which provides the generated menu to the user. The correspondence between each unit and the device or control unit is not limited to the above examples and various changes can be made.

[0614] In the above embodiments, examples of specific processing 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.

[0615] Furthermore, the emotion-specific model 59, acting as an emotion engine, can determine a user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine a user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The emotion-specific model 59 can also determine the robot's emotion, and the specific processing unit 290 performs specific processing based on the robot's emotions.

[0616] Figure 9 This is a diagram representing an emotion map 400 that maps multiple emotions. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. The closer to the center of the concentric circles, the more primitive the emotion is. On the outer side of the concentric circles, emotions representing states or behaviors arising from mood are arranged. Emotions are concepts that include feelings and mental states. Emotions generated by reactions occurring in the brain are arranged roughly to the left of the concentric circles. Emotions derived from situational judgments are arranged roughly to the right of the concentric circles. Emotions generated by reactions occurring in the brain and derived from situational judgments are arranged roughly above and below the concentric circles. Furthermore, "pleasant" emotions are arranged above the concentric circles, and "unpleasant" emotions are arranged below them. Thus, in the emotion map 400, multiple emotions are mapped based on the structure that generates emotions, and emotions that are likely to occur simultaneously are mapped close to each other.

[0617] These emotions are distributed at the three o'clock position of the emotion map 400, typically fluctuating between peace and anxiety. In the right half of the emotion map 400, situational awareness dominates over internal sensation, thus resulting in an impression of calm.

[0618] The inner side of the emotion map 400 represents the inner state, while the outer side represents behavior. Therefore, the further outward you are from the emotion map 400, the more visible the emotion becomes (manifested in behavior).

[0619] Here, human emotions are based on various balances such as posture and blood sugar levels. When these balances deviate from an ideal state, it indicates an unpleasant state; when they approach the ideal state, it indicates a pleasant state. Emotions in robots, cars, motorcycles, etc., can also be created in the following way: based on various balances such as posture and remaining battery power, when these balances deviate from an ideal state, it indicates an unpleasant state; when they approach the ideal state, it indicates a pleasant state. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on a Brain Physiological Signal Analysis System for Voice Emotion Recognition and Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the sensory-dominated region, called "response," are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the situational cognition-dominated region, called "situation," are arranged.

[0620] In the emotion map, two types of emotions that promote learning are defined. One is a negative emotion on the situational side, in the middle or peripheral region of "repentance" or "reflection." This occurs when the robot experiences negative emotions such as "I don't want to experience this feeling again" or "I don't want to be blamed again." The other is a positive emotion on the response side, near the "desire" region. This occurs when there are positive feelings such as "wanting more" or "wanting to know more."

[0621] The emotion-specific model 59 inputs user input into a pre-trained neural network to obtain emotion values ​​representing each emotion shown in the emotion map 400, thereby determining the user's emotion. This neural network is pre-trained based on multiple training data sets, which are combinations of user input and emotion values ​​representing each emotion shown in the emotion map 400. Furthermore, this neural network... Figure 10 As shown in the sentiment graph 900, it was trained in a way that sentiments that are configured close to each other have similar values. Figure 10 The text shows examples of emotions such as "peace of mind", "stability", and "reassurance" that have similar emotion values.

[0622] The above description focuses on the functions of the data processing device 12, but the system of this disclosure is not necessarily installed on a server. The system of this disclosure can also be installed as a general information processing system. This disclosure can also be installed, for example, as a software program running on a personal computer, an application running on a smartphone, etc. The method of this disclosure can also be provided to users in the form of SaaS (Software as a Service).

[0623] In the above embodiments, an example of a specific process being performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and the specific process can also be distributed among multiple computers, including computer 22. For example, the data generation model 58 can be located on an external device of the data processing apparatus 12, where data is generated based on the input data.

[0624] In the above embodiments, examples of storing a specific processing program 56 in the memory 32 have been described, 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 into the computer 22 of the data processing device 12. The processor 28 executes specific processing according to the specific processing program 56.

[0625] Alternatively, a specific processing program 56 may be pre-stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 according to the requirements of the data processing device 12.

[0626] In addition, it is not necessary to store all the specific processing program 56 in the storage device such as the server connected to the data processing device 12 via the network 54 or in the memory 32; a portion of the specific processing program 56 may be stored in advance.

[0627] As hardware resources for performing specific processes, various processors, as shown below, can be used. For example, a CPU can be listed as a processor, which functions as a general-purpose processor that performs specific processes by executing software, i.e., a program. Furthermore, processors can be listed as special-purpose circuits such as FPGAs (Field-Programmable Gate Arrays), PLDs (Programmable Logic Devices), or ASICs (Application-Specific Integrated Circuits), which are processors with circuitry specifically designed to perform specific processes. Each processor has built-in or connected memory, and each processor executes specific processes using that memory.

[0628] 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). Alternatively, the hardware resources for performing a specific process can be a single processor.

[0629] As an example of a single processor, there are two approaches: First, a processor is composed of a combination of one or more CPUs and software, which functions as a hardware resource to perform a specific process; second, as represented by a SoC (System-on-a-chip), a processor is used to implement the functionality of the entire system, which includes multiple hardware resources for performing a specific process, using a single IC (Integrated Circuit) chip. In this way, the specific process is implemented by using one or more of the aforementioned processors as hardware resources.

[0630] Furthermore, the hardware architecture of these various processors, more specifically, can utilize circuits that combine semiconductor elements and other circuit components. Moreover, the specific process described above is just one example. Therefore, without departing from the main point, unnecessary steps can certainly be deleted, new steps added, or the processing order changed.

[0631] The descriptions and illustrations above are detailed explanations of a portion of the technology disclosed herein, and are merely one example of the technology disclosed herein. For example, the above descriptions of the structure, function, effect, and results are just one example of the structure, function, effect, and results of a portion of the technology disclosed herein. Therefore, without departing from the spirit of the technology disclosed herein, unnecessary parts may be deleted, new elements added, or replacements may be made to the descriptions and illustrations above. Furthermore, to avoid confusion and facilitate understanding of a portion of the technology disclosed herein, explanations of common technical knowledge that do not require special explanation under the premise of being able to implement the technology disclosed herein have been omitted from the descriptions and illustrations above.

[0632] All documents, patent applications and technical specifications set forth in this specification are incorporated herein by reference to the same extent that each document, patent application and technical specification is specifically and individually described therein and referenced by reference.

[0633] In addition, the following notes are provided in response to the above explanation.

[0634] Example 1 (Note 1) An information processing system, characterized in that it comprises: An apparatus for acquiring structured data from public and private information sources, and using a relational database or equivalent data management function to set and store attribute unification and consistency constraints on the structured data, and to distinguish between population-related data and budget-related data from public information sources and behavior-related data or market-related data from private information sources, and to generate a unified data set using the same region and the same year as keys. An apparatus for using analysis software with tabular data processing capabilities to calculate, using the integrated dataset, at least several evaluation indicators including population growth rate, budget utilization efficiency, and correlation indicators between these indicators and private sector indicators, to perform statistical analysis along the time and geographical directions, to set evaluation benchmarks or target values ​​for social capital construction based on the calculated evaluation indicators, and to store the evaluation benchmarks or target values ​​as indicator information in a database. An apparatus for automatically or semi-automatically generating prompts for inputting into a generative artificial intelligence model by using the indicator information and the summary results or summary information extracted from the integrated data set as background information, providing the prompts and the background information to the generative artificial intelligence model to obtain descriptive information including multi-year funding plans, budget allocation schemes and performance evaluations, and automatically generating application documents, report documents and accompanying tables or graphs in a predetermined format based on the obtained descriptive information, indicator information and the integrated data set; A means for storing and / or outputting the automatically generated application documents and report documents in electronic file form to a display device.

[0635] (Note 2) The information processing system according to Appendix 1 is characterized in that, The device for generating prompt statements is configured to: extract regions from the integrated data set whose population growth rate and budget utilization efficiency are lower than the evaluation benchmark or target value on a regional basis, and attach the specific results as constraints and output formats to the prompt statements for specific improvement areas and priority investment areas for each region, so that the generative artificial intelligence model outputs optimization suggestions for each region in tabular or columnar form.

[0636] (Note 3) The information processing system according to Appendix 1 is characterized in that, The system also includes a device for calculating trend information and future prediction intervals for multiple time periods based on the integrated data set and the evaluation indicators, and including the trend information and future prediction intervals in the prompt statement to instruct the generative artificial intelligence model to generate multi-year social capital construction plans, risk factors and countermeasures, and automatically embedding the output of the generative artificial intelligence model as part of the application document and the report document.

[0637] Application Example 1 (Note 1) An information processing system, characterized in that it comprises: An apparatus for acquiring publicly available information from public institutions and information held by private entities through communication means, aligning the information based on time and location information, and storing the aligned information as fused data. An apparatus for extracting multiple attribute information, including population information, behavioral information, event occurrence information and financial information, from the fused data, and performing statistical processing and preprocessing on the attribute information to generate a learning dataset and a reasoning dataset. An apparatus for constructing a machine learning model based on the learning dataset, calculating the probability of events occurring in a spatiotemporal region based on the inference dataset, and calculating countermeasure indicators including patrol behavior indicators, monitoring device configuration indicators, and fund allocation indicators based on the event occurrence probability. An apparatus for calculating an evaluation benchmark, including a countermeasure indicator value, an evaluation indicator value, and a fund utilization indicator value, based on the countermeasure indicator and the fused data. A device for generating the evaluation benchmark and the countermeasure indicators into machine-readable structured data, and generating prompt statements that include the structured data as input information. A device for inputting the prompt statement and the structured data into a generative artificial intelligence model, and using the natural language text obtained from the generative artificial intelligence model to automatically generate document information conforming to a predetermined format, the document information including report information, application information and supplementary information; An apparatus for generating notification information based on at least one of the countermeasure indicators and the document information, and sending the notification information to an external terminal device; A device for updating the fused data and the machine learning model based on implementation result information obtained from the external terminal device, and for resetting at least one of the prompt statement and the evaluation benchmark.

[0638] (Note 2) According to the information processing system described in Appendix 1, the device for calculating countermeasure indicators is configured to combine information from public institutions specifically corresponding to a particular administrative region with information from private entities corresponding to that administrative region in the fused data, estimate the probability of events occurring in the administrative region based on the combination, and generate region-specific countermeasure indicators for the administrative region, including the patrol behavior indicator, the monitoring device configuration indicator, and the fund allocation indicator.

[0639] (Note 3) According to the information processing system described in Appendix 1, the device for generating prompt statements is configured to generate multiple prompt statements based on predetermined format information and user attribute information, including control information for specifying the output format, style, level of detail, and key items to the generative artificial intelligence model, and to adjust the information content, expression, and structure of the natural language text obtained from the generative artificial intelligence model according to the user attribute information, so as to control the output result of the generative artificial intelligence model.

[0640] Example 2 (Note 1) An information processing system, characterized in that it comprises: An apparatus for communicating with multiple information sources containing public and private information, obtaining the public and private information using structured queries, and integrating the obtained information as a single data set; An apparatus for generating structured data corresponding to each item in a document template by using a general program execution environment and data parsing software to perform missing data processing, type conversion, summary calculation and evaluation index calculation on the integrated data set. An apparatus for generating tables and graphs using visualization software based on the integrated data set and the evaluation indicators, and for associating the tables and graphs with predetermined positions within the document template; An apparatus for automatically generating prompt statements for input to a generative artificial intelligence model based on the structured data and the document template, providing the prompt statements and a summary of the structured data to the generative artificial intelligence model to generate explanatory text, and arranging the explanatory text in a predetermined description area within the document template. An apparatus for generating a predetermined electronic file format for a report and a document for an application based on structured data, tables, graphics and explanatory text arranged in the document template, and for saving the electronic file to an information storage device and providing it in a state accessible by an external device.

[0641] (Note 2) The information processing system according to Appendix 1 is characterized in that the system further includes: An apparatus for updating at least a portion of the descriptions in the document for reporting and the document for application based on user-edited content received from a terminal device, checking the consistency between the updated descriptions and the integrated data set, identifying elements of inconsistency and generating warning messages, and saving a final electronic file reflecting the updated descriptions to the information storage device.

[0642] (Note 3) The information processing system according to Appendix 1 is characterized in that the system further includes: A device for modifying the content of prompts for a generative artificial intelligence model based on user requirements input from a terminal device, comparing the time, region, and quantity information contained in the prompts with the integrated data set, automatically verifying the accuracy of corresponding information in the explanatory text output from the generative artificial intelligence model, and deciding whether to adopt the explanatory text or instructing it to be regenerated based on the verification results.

[0643] Application Example 2 (Note 1) An information processing system, characterized in that it comprises: A means for obtaining structured and unstructured data from multiple public and civilian information sources by an information processing device, and integrating the structured and unstructured data according to an identifier to generate an integrated data set; Means for using statistical processing and machine learning methods to calculate evaluation indicators related to social infrastructure construction, funding provision, business evaluation, or production management from the integrated dataset, and to summarize the evaluation indicators by time unit or regional unit; Emotion inference means for inferring emotional state based on user behavior information, voice information, image information or text information obtained from sensing devices or input / output devices; Means for applying data transformation and visualization techniques to the integrated dataset and the evaluation indicators to generate tabular and graphical data, and for automatically arranging the tabular and graphical data into a document template; A means of adjusting indicators to dynamically modify target values ​​or evaluation benchmarks based on the evaluation indicators and the emotional state. A means for generating prompt statements for inputting into a generative artificial intelligence model based on the integrated dataset, the evaluation indicators, the tabular data, the graphical data, and the emotional state, and using the prompt statements to enable the generative artificial intelligence model to perform natural language generation processing, thereby automatically generating application documents, report documents, or supplementary information. A means for acquiring operational or transactional data from production equipment or information terminals in chronological order, performing missing data completion, outlier deletion, and indicator calculation on the operational or transactional data, and associating natural language descriptions generated by the generative artificial intelligence model with the visualization results of the operational or transactional data and outputting them to a user terminal.

[0644] (Note 2) The information processing system according to Appendix 1 is characterized in that, The emotion inference method is used to determine the user's stress state, anxiety state, or positive emotional state based on the feature quantities extracted from the voice information and the image information. The indicator adjustment method is used to increase the target value of the evaluation indicator when the user is determined to be in the positive emotional state, and relax the target value or simplify the content of the application document or the report document when the user is determined to be in the stress state or the anxiety state, thereby changing the prompt statements for the generative artificial intelligence model.

[0645] (Note 3) The information processing system according to Appendix 1 is characterized in that, The information processing device is used to extract crime occurrence information, security-related information, or infrastructure operation information from the integrated data set based on location information obtained by location acquisition means, and generate notification prompt statements containing warning or reassurance information according to the extraction results and the emotional state. The notification prompt statements are used to enable the generative artificial intelligence model to generate natural language notification text, and the natural language notification text is associated with geographic information display and provided to the user terminal.

Claims

1. An information processing system, characterized in that, Includes a processor, the processor being configured to: Acquire and integrate open and non-governmental data from autonomous regions; Calculate the evaluation benchmark using the integrated data; Using generative artificial intelligence models, reports can be automatically generated based on prompts.

2. The information processing system according to claim 1, characterized in that, The processor is configured to combine data from a specific autonomous region with specific civilian data to enable analysis targeting a specific region.

3. The information processing system according to claim 1, characterized in that, The processor is configured to generate reports and supplementary information in a specific format and to adjust the information according to the user's mood, for which prompts are used to indicate adjustments to the generative artificial intelligence model, thereby adjusting the output.

Images