GeoCogent

<b>GeoCogent Prototype System and GeoCodes Dataset: Overview</b><b>Introduction</b>The GeoCogent Prototype System and GeoCodes Dataset are two essential components developed to advance research and practical applications in the field of geospatial code generation. These resources provide a robust, open-access platform and evaluation framework for researchers and developers, facilitating efficient and accurate geospatial code generation while addressing core challenges such as complex demand parsing, syntax adaptation, path retrieval, and code validation.<b>GeoCogent Prototype System</b>The GeoCogent Prototype System is an interactive, user-friendly tool designed to automate geospatial code generation using a large language model (LLM)-driven agent framework. The system integrates three key mechanisms—planning, tools, and memory—to enable dynamic interaction with user inputs and external environments.<b>Key Features</b><br><b>Local Deployment</b>: Run the system locally by executing <code>app.py</code> for ease of use and flexibility.<b>Knowledge Integration</b>: Manage knowledge bases via JSON files and task data through TXT files for seamless operation.<b>LLM Integration</b>: Supports the connection and extension of various LLMs to enhance geospatial code generation tasks.<b>User Accessibility</b>: Offers an intuitive interface and detailed user manual, enabling researchers and developers from various fields to leverage geospatial code generation tools without extensive technical expertise.<b>GeoCodes Dataset</b>GeoCodes is a comprehensive benchmark dataset specifically designed for evaluating the performance of geospatial code generation models. It consists of 50 diverse tasks across three demand types—explicit demands, incomplete demands, and open-ended demands—encompassing various platforms, programming languages, data sources, and output types.<b>Key Characteristics</b><br><b>Diverse Task Coverage</b>: Tasks range from straightforward geospatial analyses to more complex, context-specific operations.<b>Standardized Reference Materials</b>: Each task includes demand parsing documents, executable code, and reference outputs to ensure consistency in evaluation.<b>Open-Access</b>: GeoCodes is freely available to the research community, providing a standard framework for performance comparisons across geospatial code generation models.<b>Supplementary Materials</b>To assist readers in understanding the figures and tables presented in the manuscript, we provide a dedicated <code>Code/</code> folder as supplementary material. This folder includes:Script references and explanatory notes related to <b>Table 3–5</b>, clearly specifying the input-output mappings for each evaluation setting.A figure-to-file mapping table for <b>Figure 7–13</b>, linking each visual result to its associated task case and script path within the <code>GeoCodes/</code> dataset.A detailed <code>README.md</code> file that explains how each script relates to the manuscript content.<b>File Structure Overview</b>├── Code/<br>│ ├── app.py # Main entry point for prototype system (Figure 13)<br>│ ├── environment.py # Environment mechanism implementation<br>│ ├── memory.py # Memory mechanism implementation<br>│ ├── tool.py # Tool mechanism implementation<br>│ ├── eq1_generate_code_batch.py # EQ1 experiments → Table 3<br>│ ├── eq2_generate_code_batch.py # EQ2 experiments → Table 4<br>│ ├── eq3_generate_code_batch.py # EQ3 experiments → Table 5<br>│ ├── index.html # System homepage interface<br>│ ├── script.js / styles.css / logo.png # UI assets <b>Impact and Contribution</b>Together, the GeoCogent Prototype System and GeoCodes Dataset aim to:<br>Lower technical barriers for geospatial code development.Provide a standardized evaluation framework to measure the effectiveness of geospatial code generation models.Facilitate cross-disciplinary collaboration and innovation in geospatial analytics.Support researchers and practitioners with practical tools for real-world geospatial applications.<br>