Multi-modal evaluation report generation method and system
By using a multimodal assessment report generation method, which automatically parses planning texts using NLP and knowledge graphs, and combines machine learning clustering algorithms to discover urban functional zones, interactive reports are dynamically generated. This solves the problems of low efficiency, insufficient scientific rigor, and strong subjectivity in urban development planning assessment, and achieves efficient and scientific assessment results and a global insight.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-09
- Publication Date
- 2026-03-13
AI Technical Summary
Existing technologies for evaluating urban development plans suffer from problems such as inefficient and inconsistent indicator formulation, disconnect between data and text, rigid and difficult-to-update evaluation reports, and strong subjectivity in the evaluation process, resulting in insufficient scientific rigor and credibility of the evaluation.
A multimodal assessment report generation method is adopted, which uses NLP and knowledge graph to automatically parse planning text, combines machine learning clustering algorithm to discover urban functional areas, dynamically generates interactive reports, and improves the scientific nature of the assessment through a hybrid weight optimization algorithm.
It has achieved an exponential improvement in assessment efficiency, significantly enhanced the scientific rigor and credibility of assessment results, and significantly improved the interactivity and reproducibility of reports, providing a holistic insight into "planning vs. reality" and supporting dynamic monitoring and agile decision-making.
Smart Images
Figure CN121660503A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of development planning, data analysis and decision support technology, and in particular relates to a method and system for generating multimodal assessment reports. Background Technology
[0002] Development plans are guiding documents for a city's future development. In the mid-to-late stages of implementation, conducting scientific, comprehensive, and timely evaluations of their effectiveness is crucial for ensuring the achievement of planning goals and revising the development path. Currently, the evaluation of development plans typically relies on the following traditional technical methods: 1. Manual Interpretation and Indicator Development: The primary task of planning evaluation is to develop a quantitative evaluation indicator system based on the planning documents. This process typically involves planning experts and government staff manually reading, discussing, and interpreting numerous policy documents, extracting key objectives, and transforming them into quantifiable evaluation indicators. 2. Manual Data Collection from Multiple Sources: After determining the indicators, evaluators need to manually collect supporting data from different departments and systems. For example, calculating "park coverage" requires obtaining spatial data on parks from the planning department, resident population data from the statistics department, and data on the built-up area from the planning and resources bureau. The data formats vary, requiring extensive manual cleaning and alignment. 3. Static Report Writing: The evaluation results are ultimately presented in report form, usually as a Word or PDF document. Writing the report is a tedious "copy-paste" process; evaluators need to manually stitch together statistical tables from Excel, screenshots from GIS software, and charts from other analytical tools into the report document. 4. Subjective Weight Allocation: When conducting comprehensive evaluations, the importance (i.e., weight) of different indicators is often determined through methods such as expert scoring or the analytic hierarchy process (AHP). This process is highly subjective and relies on the experience and preferences of experts, which may lead to questions about the scientific validity and credibility of the evaluation results.
[0003] These existing technologies have revealed significant shortcomings in practice: 1) Inefficient and inconsistent indicator formulation: Manual indicator formulation is not only time-consuming and labor-intensive, but also prone to discrepancies in understanding of the same policy text among different evaluators, leading to a lack of consistency and reproducibility in the developed indicator system. 2) Severe disconnect between data and text: There is a lack of automated correlation between the objectives in the planning text and the underlying data used for calculation, making data retrieval difficult and failing to guarantee complete alignment between the indicator calculation methods and the original policy intent. 3) Rigid evaluation report format and difficulty in updating: Once a static report is generated, interactive exploration (such as drill-down and filtering) is impossible. When the source data is updated, the entire report requires significant effort to recreate, failing to meet the needs of dynamic monitoring and agile decision-making. 4) High subjectivity and insufficient scientific rigor in the evaluation process: The allocation of weights relies too heavily on subjective judgment and lacks data-driven objective basis, significantly reducing the scientific validity and persuasiveness of the evaluation conclusions and making it difficult to fully support complex decision-making scenarios. Summary of the Invention
[0004] The purpose of this invention is to provide a method and system for generating multimodal assessment reports. This method can automatically generate assessment indicators precisely linked to spatial data from planning texts, dynamically generate interactive multimodal assessment reports, and employ a more scientific algorithm to optimize indicator weights, thereby fundamentally transforming the assessment model of urban development planning. The technical solution adopted is as follows: A method for generating a multimodal evaluation report includes the following steps: Step 1: Convert the planning text into an evaluation indicator system and quantify the evaluation indicators to obtain evaluation indicator values; Step 2: Assign combined weights to each evaluation indicator; Step 3: Mark the actual functional zones of the city to obtain an urban functional zone map that includes several actual functional zones of the city; Step 4: Integrate the evaluation index values, evaluation indicators, combined weights, and the actual functional zoning map of the city to generate a multimodal evaluation report.
[0005] Preferably, step 1 specifically includes the following steps: Step 1.1: Parse the planning text and identify and extract entities, indicator terms, and constraints; Step 1.2: Construct a knowledge graph based on the extracted entities and key terms; Knowledge graphs take entities and constraints as inputs and output evaluation metrics as outputs. Step 1.3: Automatic generation of evaluation indicators and binding with quantitative formulas.
[0006] Preferably, step 2 specifically includes the following steps: Step 2.1: Calculate the objective weights; Step 2.2: Collection of expert experience weights; Step 2.3: Weight the expert experience weight and the objective weight to obtain the combined weight.
[0007] Preferably, step 3 specifically includes the following steps: Spatial unit division and feature construction, K-Means clustering analysis, and semantic interpretation and automatic annotation of functional area clusters.
[0008] Preferably, step 4 specifically includes the following steps: Step 4.1, Data-driven dynamic rendering: Step 4.2: Dynamic component integration and interactive report output.
[0009] A multimodal evaluation report generation system, comprising a method for generating multimodal evaluation reports, including: The intelligent indicator generation module is used to convert planning text into an evaluation indicator system and quantify the evaluation indicators to obtain evaluation indicator values. The dynamic weight optimization module is used to generate combined weights; The urban functional zone discovery module is used to identify the actual functional zones of a city. The multimodal report generation module is used to dynamically fuse multi-source data into interactive reports; The multimodal report generation module is connected to the intelligent indicator generation module, the dynamic weight optimization module, and the urban functional area discovery module at its input end.
[0010] In summary, this invention proposes a closed-loop intelligent evaluation framework encompassing "text-indicators-data-report". The core of this framework is to take unstructured planning policy texts as input, and through a series of automated models, ultimately output a structured, data-driven, interactive, and scientifically weighted multimodal evaluation report. Its implementation path includes four core innovations.
[0011] (1) Intelligent generation of indicators: Using NLP and knowledge graph technology, the planning text is automatically parsed, key objectives are identified, and they are intelligently mapped into quantitative evaluation indicators that are precisely bound to spatial data entities.
[0012] (2) Dynamic weight optimization: Combining the objective weighting method based on data information entropy with expert experience, a hybrid weight model is formed to improve the scientific nature of the comprehensive evaluation.
[0013] (3) Unsupervised discovery of urban functional areas: Machine learning clustering algorithms are introduced to conduct unsupervised learning on big data of urban points of interest, and automatically discover the objectively existing and spontaneously formed functional areas (such as commercial areas, residential areas, industrial areas, etc.) in urban space, providing a dynamic and data-driven realistic baseline for planning evaluation.
[0014] (4) Dynamic Report Generation: Using a template engine, the following multiple modal contents are dynamically integrated into a standardized and interactive report: GIS map service (steps 3 and 4), interactive charts (step 4), structured data (steps 1, 2, and 4), and analytical text (step 4).
[0015] Compared with the prior art, the advantages of the present invention are: 1. Exponential improvement in evaluation efficiency: Intelligent generation of indicators is achieved through NLP and knowledge graphs, which reduces the work that used to take weeks of manual discussion and formulation to minutes of automated processing, greatly freeing up manpower.
[0016] 2. Fundamental enhancement of the scientific rigor of the assessment: The innovative hybrid weight optimization algorithm combines data-driven objective laws with valuable expert experience, overcoming the bias of purely subjective weighting and making the comprehensive assessment conclusions more scientific, fair and credible.
[0017] 3. Achieved deep integration of text, data and space: This invention establishes a strong and automated link between the macro-level objectives of planning texts and the micro-level spatial entities of GIS through knowledge graphs and data standards, ensuring a high degree of consistency between assessment criteria and policy intentions.
[0018] 4. Providing a holistic insight into "planning vs. reality": This invention's unique unsupervised urban functional area discovery module can reveal the spontaneously formed true functional structure of the city from objective data. This allows the system to go beyond simply "verifying" the implementation of plans, proactively "discovering" deviations between plans and reality, providing decision-makers with unprecedented strategic insights and dynamic adjustment basis, achieving a qualitative leap from passive monitoring to proactive analysis.
[0019] 5. Revolutionary Interactive Report Experience: Revolutionizing the traditional static and fragmented report format, it creates a "living report" that integrates maps, charts, tables, and text. Users can directly explore data within the report, greatly improving the intuitiveness of information acquisition and decision-making efficiency.
[0020] 6. Strong standardization and reproducibility: The automated process ensures that each assessment follows the same standards and methods, and the assessment results are highly reproducible and comparable across different regions, laying a technical foundation for establishing a long-term and standardized planning assessment mechanism. Attached Figure Description
[0021] Figure 1 The system architecture diagram upon which the multimodal evaluation report generation method is based; Figure 2 Flowchart for generating multimodal evaluation reports. Detailed Implementation
[0022] The multimodal evaluation report generation method and system of the present invention will be described in more detail below with reference to the schematic diagrams, which illustrate preferred embodiments of the present invention. It should be understood that those skilled in the art can modify the present invention described herein while still achieving the advantageous effects of the present invention. Therefore, the following description should be understood as being of general knowledge to those skilled in the art and is not intended to limit the present invention.
[0023] like Figures 1-2 Methods for generating multimodal evaluation reports include: Step 1: Convert the planning text into an evaluation indicator system and quantify the evaluation indicators to obtain evaluation indicator values.
[0024] The goal is to automatically convert unstructured planning texts into a structured, computable evaluation index system.
[0025] Specifically, evaluation metrics are intelligently generated based on NLP and knowledge graphs.
[0026] Step 1.1: Parse the planning text and identify and extract entities, indicator terms, and constraints.
[0027] First, the system receives the input planning policy document (planning text). Then, it uses Named Entity Recognition (NER) technology in Natural Language Processing (NLP) to perform deep analysis on the planning text, automatically identifying and extracting predefined key entity categories.
[0028] Entity categories include target entities and spatial entities.
[0029] The target entity includes all housing in the city.
[0030] Spatial entities (Location) include: administrative districts, key areas, and parks.
[0031] Metric keywords are evaluation indicators, including: coverage, accessibility, area per capita, proportion, quantity, etc.
[0032] Constraints: "higher than" or "not lower than".
[0033] Step 1.2: Construct a knowledge graph based on the extracted entities and index terms.
[0034] Knowledge Graph: A knowledge graph is a directed graph network structure consisting of nodes (entities, metrics, concepts) and directed edges (relationships). Its input is a triple of {entity, metric, constraint}, and its output is a quantifiable evaluation metric.
[0035] Construction method: A rule-based construction method is adopted. The construction process includes: (a) predefining the entity library, indicator library, and relation library of the planning domain; (b) querying the corresponding relations and transformation rules in the library based on the entity and indicator terms extracted in step 1.1; (c) using the constraints as parameters, generating the final evaluation indicator definition through relations and rules.
[0036] This knowledge graph stores knowledge in the form of "entity-relationship-entity" triples.
[0037] The meaning of a triplet: The first entity: the target entity identified in the planning text (such as "park" or "housing"). Relationship: The corresponding logical relationship between the entity and the evaluation indicators (such as "has", "services", "supply", etc.). The second entity / metric term: Quantitative metric terms (such as "coverage", "quantity", "accessibility", etc.) Example: Input: Entity = Park, Indicator = Coverage, Constraint = "≥80% within a service radius of 500 meters" The knowledge graph query yielded the relationship: "Park-Service-Coverage". Output evaluation metric: Service coverage rate within 500 meters of the park ≥ 80% The core value of this knowledge graph lies in the fact that it not only establishes a connection between textual goals and abstract indicators, but also precisely binds abstract indicators to specific, callable GIS spatial data layers and services by connecting to data standards.
[0038] Step 1.3: Automatic generation of evaluation indicators and binding with quantitative formulas.
[0039] First, based on the entities extracted from the planning text, reasoning is performed on the knowledge graph to automatically generate evaluation indicators.
[0040] Then, each evaluation indicator is bound to the GIS data layer and quantitative calculation formula required for its calculation.
[0041] GIS data layer binding: Specifies the GIS spatial data source required to calculate this indicator. For example, the "Park Coverage" indicator can be bound to the "Park POI Data Layer" and the "Built Area Boundary Layer". Calculation formula template: Defines a parameterized calculation formula template for this indicator, but does not perform the calculation in this step. For example, "Park coverage rate = Number of grids covered by parks / Total number of grids".
[0042] The binding process described above is based on existing technologies (data dictionary, metadata management, etc.). The innovation of this step lies in accelerating metric generation from manual formulation to automatic derivation through the automation of knowledge graph reasoning. The actual metric value calculation occurs in step 4 (during report generation), at which point the system calls real-time data to complete the calculation.
[0043] For example, once the entity "park" is identified, the knowledge graph is queried to automatically generate a series of related evaluation indicators such as "number of parks", "park service radius coverage", and "per capita park green space area".
[0044] Step 2: Assign combined weights to each evaluation indicator.
[0045] This step aims to address the subjectivity of weighting evaluation indicators by introducing a data-driven scientific approach.
[0046] Step 2.1, Objective weight calculation (based on entropy weight method): For a given evaluation scenario, the system automatically collects data for all indicators within that scenario at different units or time points. The entropy weighting method is used to calculate the objective weights. The basic principle of this method is: the greater the difference in an indicator's data value across evaluation units (i.e., the smaller the information entropy), the more effective information the indicator provides, and its weight should be higher. Conversely, if all units score similarly on the indicator, its weight should be lower. The system automatically performs this calculation, obtaining a set of objective weights.
[0047] Step 2.2, Expert Experience Weighting: The system provides a simple online interface for planning experts or managers to subjectively score the same set of indicators based on their policy importance and experience (e.g., on a 1-5 scale). The system then normalizes the scores to obtain a set of subjective weights.
[0048] Step 2.3: Weight the expert experience weight and the objective weight to obtain the combined weight.
[0049] Step 3: Mark the actual functional zones of the city to obtain an urban functional zone map that includes several actual functional zones of the city.
[0050] This module is one of the core innovations of this invention, aiming to automatically "discover" the actual functional structure of a city from massive geospatial data, providing a dynamic, data-driven real-world context for planning and evaluation.
[0051] (1) Spatial unit division and feature construction: First, the target urban area is divided into regular grids of uniform size (e.g., 500m x 500m) as the basic analysis unit for functional zone identification.
[0052] Then, for each grid cell, a high-dimensional functional feature vector is constructed based on the points of interest (POI) data contained within it.
[0053] Each dimension of the vector represents the density or frequency of a specific POI category within the grid.
[0054] For example, a vector can be represented as V = [V_Business, V_Office, V_Residential, V_Public Services,...], where the value of each element is the number or density of POIs in the corresponding category.
[0055] (2) K-Means clustering analysis: The K-Means clustering algorithm is applied by taking the set of feature vectors of all grid cells as input.
[0056] This algorithm uses iterative calculations to group grid cells with similar functional structures (i.e., feature vectors that are close in distance in multidimensional space) into the same cluster.
[0057] Meanwhile, methods such as the Elbow Method or Silhouette Score can be used to help determine the optimal number of clusters K, ensuring the rationality of the clustering results.
[0058] (3) Semantic interpretation and automatic annotation of functional area clusters: After clustering is completed, the system interprets the macroscopic function of each cluster by analyzing the centroid of each cluster, which is the average vector of all grid feature vectors within that cluster.
[0059] The dimension with the highest value in the centroid vector reflects the dominant functional type of this cluster.
[0060] Based on the centroid analysis results, automatic labeling is performed using a set of preset rules.
[0061] For example, if the centroid of a cluster is significantly higher in the dimensions of "commercial retail" and "corporate office" than in other dimensions, then the cluster is automatically labeled as "central business district". If the centroid of a cluster is absolutely dominant in the "living" dimension, it is labeled as a "residential area".
[0062] Step 4: Dynamically generate multimodal evaluation reports based on template engine.
[0063] This step aims to automate, visualize, and interactively present the analysis results.
[0064] Step 4.1, Data-driven dynamic rendering: When a user initiates a request to generate an evaluation report, the system triggers the report generation framework.
[0065] The framework automatically retrieves the output of the preceding steps based on the request: The evaluation indicators related to the entity and their associated data are obtained from the indicator generation module. In this example, the evaluation indicators related to the entity "housing" are obtained.
[0066] Obtain the mixed weights of relevant indicators from the weight optimization module.
[0067] Obtain the latest actual urban functional area layer from the urban functional area discovery module.
[0068] Obtain the necessary GIS data and business statistics from the data platform, including administrative boundaries, POI locations, building vectors, population thermal grids, and transportation networks.
[0069] Among them, business statistics data include non-spatial data such as population statistics, economic data, and project completion numbers from various government departments.
[0070] Data platform: This is the underlying support module that provides data services to all upper-level modules.
[0071] Step 4.2: Dynamic component integration and interactive report output.
[0072] Dynamic component integration: Integrate the outputs of multiple dynamic components into a unified HTML5 report page through a template engine.
[0073] Dynamic components: Front-end elements that can be dynamically rendered based on actual data, including maps, charts, tables, and text.
[0074] The template engine starts working, replacing the placeholders in the template with the retrieved real data: Call the GIS map service, pass in the data and rendering parameters, generate an interactive housing density heat map, and embed the report.
[0075] The data transmitted to the GIS map service includes: Evaluation indicator values (such as the number of housing units in each grid). Urban functional zone layer; Reference boundaries such as built-up areas; Rendering parameters such as color grading and transparency.
[0076] Meanwhile, the actual urban functional zoning map (real layout) generated in step 3 can be used as an overlay layer to achieve an intuitive comparison between the planned layout and the actual pattern.
[0077] The planning layout is output from step 1, which requires extracting spatial planning intent or planning drawings from the planning text.
[0078] Call the ECharts chart library, pass in supply and demand data, generate an interactive bar chart, and embed it in the report.
[0079] Fill in the cost-benefit analysis data in the table.
[0080] Ultimately, the system outputs not a static PDF, but a single, self-contained HTML5 page.
[0081] Users can directly zoom in and out of the GIS map service, hover over to view chart data, and click to sort tables within this "report," achieving a brand-new experience where the report is an application.
[0082] like Figure 1 As shown, this system mainly consists of the following six cooperating modules: Intelligent indicator generation module: Responsible for implementing step 1. It contains an NLP parsing engine and a knowledge graph database. Its core function is to automatically generate quantitative indicators that are bound to spatial data from planning text.
[0083] Dynamic weight optimization module: Responsible for implementing step 2. It includes the core of entropy weight calculation and expert weight input interface, used to generate scientific hybrid evaluation weights.
[0084] Urban Functional Zone Discovery Module: Responsible for implementing step 3. It contains a POI data processing engine and the core of the K-Means clustering algorithm, used to automatically mine actual urban functional zones from big data.
[0085] Multimodal report generation module: Responsible for implementing step 4. Its core components are the template engine and data scheduler, which dynamically fuse multi-source heterogeneous data into interactive reports.
[0086] Data Management and Service Module: As the underlying support, it is responsible for connecting internal and external data sources and providing unified and standardized data interface services for upper-layer modules.
[0087] Visualization and Interaction Module: As the user front end of the system, it provides an interface for indicator management, weight adjustment, report generation requests, and the final interactive report presentation.
[0088] The multimodal report generation module is connected to the intelligent indicator generation module, the dynamic weight optimization module, and the urban functional area discovery module at its input end.
[0089] Figure 1 In the middle: Data Management and Service Module: Provides raw data and a unified data interface for all core processing layer modules (especially the intelligent indicator generation module, dynamic weight optimization module, and urban functional area discovery module). Knowledge graph: mainly supports the "intelligent indicator generation module" (used to infer evaluation indicators and their constraints from planning text). GIS map service: mainly supports the "Urban Functional Zone Discovery Module" and the "Multimodal Report Generation Module" (spatial analysis and functional zone division results are rendered into the report).
[0090] EChart Chart Library: Primarily supports the "Multimodal Report Generation Module" (generating various charts as part of the report content).
[0091] For some core aspects of the technical solution of this invention, there are feasible technical alternatives that can be applied to different levels of technology maturity, cost budgets, or specific business needs: 1. Alternative solutions for the intelligent indicator generation process: The Large Language Model (LLM)-based approach avoids constructing complex knowledge graphs and instead employs a prompting engineering method based on large language models (such as the GPT series and Wenxin Yiyan). By designing cleverly crafted prompts, the large model directly reads the planning text and is required to output suggestions in a specified format (such as JSON), including evaluation metrics, required data sources, and computational logic. This approach offers rapid development, but the controllability and accuracy of the results may require additional validation steps.
[0092] 2. Alternative solutions to the indicator weight optimization algorithm: Digitalization of the Analytic Hierarchy Process (AHP): For scenarios that place greater emphasis on expert experience, the traditional AHP process can be digitized and tool-based. The system guides experts to complete pairwise comparisons between indicators online and automatically calculates the consistency of the judgment matrix, ultimately deriving the weights. This is a scientific weighting method that replaces the entropy weighting method and places greater emphasis on subjective experience.
[0093] Principal Component Analysis (PCA): When there is a strong correlation between indicators, PCA can be used for dimensionality reduction, and the weights can be determined using the variance contribution rate of each principal component. This is a purely data-driven, objective weighting alternative.
[0094] 3. Alternative solutions for the multimodal report generation process: Integration with third-party business intelligence (BI) tools: Instead of developing its own report template engine, the system can be deeply integrated with mature BI tools (such as Tableau, Power BI, FineReport, etc.) via API. This system handles data processing and metric calculation, then pushes the results to the BI tool, which is responsible for rendering and presenting the final dashboard (i.e., interactive report). This approach leverages the powerful visualization capabilities of BI tools but may involve additional software licensing costs.
[0095] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.
Claims
1. A method for generating a multimodal evaluation report, characterized in that, Includes the following steps: Step 1: Convert the planning text into an evaluation indicator system and quantify the evaluation indicators to obtain evaluation indicator values; Step 2: Assign combined weights to each evaluation indicator; Step 3: Mark the actual functional zones of the city to obtain an urban functional zone map that includes several actual functional zones of the city; Step 4: Integrate the evaluation index values, evaluation indicators, combined weights, and the actual functional zoning map of the city to generate a multimodal evaluation report.
2. The multimodal evaluation report generation method according to claim 1, characterized in that, Step 1 specifically includes the following steps: Step 1.1: Parse the planning text and identify and extract entities, indicator terms, and constraints; Step 1.2: Construct a knowledge graph based on the extracted entities and key terms; Knowledge graphs take entities and constraints as inputs and output evaluation metrics as outputs. Step 1.3: Automatic generation of evaluation indicators and binding with quantitative formulas.
3. The multimodal evaluation report generation method according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2.1: Calculate the objective weights; Step 2.2: Collection of expert experience weights; Step 2.3: Weight the expert experience weight and the objective weight to obtain the combined weight.
4. The multimodal evaluation report generation method according to claim 1, characterized in that, Step 3 specifically includes the following steps: Spatial unit division and feature construction, K-Means clustering analysis, and semantic interpretation and automatic annotation of functional area clusters.
5. The multimodal evaluation report generation method according to claim 1, characterized in that, Step 4 specifically includes the following steps: Step 4.1, Data-driven dynamic rendering: Step 4.2: Dynamic component integration and interactive report output.
6. A multimodal evaluation report generation system, used to implement the multimodal evaluation report generation method according to any one of claims 1 to 5, characterized in that, include: The intelligent indicator generation module is used to convert planning text into an evaluation indicator system and quantify the evaluation indicators to obtain evaluation indicator values. The dynamic weight optimization module is used to generate combined weights; The urban functional zone discovery module is used to identify the actual functional zones of a city. The multimodal report generation module is used to dynamically fuse multi-source data into interactive reports; The multimodal report generation module is connected to the intelligent indicator generation module, the dynamic weight optimization module, and the urban functional area discovery module at its input end.