Target positioning and event tracking visualization system and method based on large language model

Abstract

The invention belongs to but not limited to the technical field of data processing, and discloses a target positioning and event tracking visualization system and method based on a large language model, and the system comprises a user input processing module which receives unstructured natural language input from a user, extracts the intention of the user, and completes the scheduling of a downstream module and the initialization of a processing flow; the big language model event extraction module based on the RAG is used for fusing the retrieval enhancement generation framework and the big language model and automatically identifying and extracting structured event information from a natural language; the geographical name analysis and space positioning module is used for standardizing geographic information entities in a natural language and accurately mapping the geographic information entities into a geographic coordinate system so as to provide support for space positioning and visualization of subsequent events; the map service and event visual display module consists of two independent services which are respectively responsible for static service of a map tile layer and visual rendering of an event track; the method is used for event position labeling, track visualization and interactive map display.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of data processing technology, and particularly relates to a target localization and event tracking visualization system and method based on a large language model. Background Technology

[0002] With the rapid development of Large Language Model (LLM) technology, it has demonstrated powerful capabilities in the field of natural language processing, especially in tasks such as information extraction, semantic understanding, and question answering systems. Leveraging its deep context modeling capabilities, LLM can extract event elements with high semantic value from unstructured text, and is widely used in scenarios requiring multi-source information integration and analysis, such as military, political, and public opinion monitoring.

[0003] In practical applications, users often want to retrieve specific event information, such as military exercises, disaster responses, or geopolitical conflicts, from massive amounts of text through natural language queries. However, traditional keyword-based retrieval methods have significant limitations in understanding context, locating spatial entities, and tracking event dynamics, making it difficult to meet users' needs for complex queries regarding the spatial distribution and temporal evolution of events.

[0004] Geographic Information Systems (GIS), as a crucial technology for spatial information processing and visualization, play a vital role in intelligence analysis, public opinion monitoring, and emergency decision-making. However, most existing GIS systems rely on structured data input and lack the ability to automatically parse and locate geographic information in natural language, making it difficult to effectively identify and visualize spatial elements implicit in text.

[0005] GeoTxt is a streaming geoparsing framework developed by a team from MITRE and Penn State University. It relies on a multi-path NER model to identify place names, then uses Apache Solr for dictionary matching and scoring within the GeoNames index, finally outputting a GeoJSON feature set. However, this solution only addresses the one-dimensional mapping of "place name → coordinates": ① It lacks structured extraction capabilities such as event trigger words and role semantics, making it unable to generate event chains that link time, space, and action; ② It completely fails to process displacement descriptions like "35 kilometers northeast of east," leaving the coordinates as static locations explicitly mentioned in the text; ③ Solr-driven inverted index retrieval experiences a significant drop in hit rate in multilingual and emerging place name scenarios, requiring manual threshold adjustment for recall-precision tradeoffs, resulting in high operational costs.

[0006] The Mordecai series employs a word vector-driven Gazetteer ranking model, parsing textual place names into GeoNames entries in Elasticsearch and providing a Docker REST interface for easy integration. While it introduces distributed vector similarity inference and improves performance on non-English corpora, it still suffers from three major technical shortcomings: ① Event granularity is limited to parsing "place name → city / country," lacking binding of direction, distance, and event semantic elements; ② Static word vectors adapt slowly to new words and administrative region changes, requiring frequent retraining; ③ The system output is pure JSON coordinates with no built-in visualization, requiring users to develop their own front-end to trace trajectories or integrate with GIS.

[0007] In summary, both existing routes remain at the "location identification + coordinate analysis" stage, failing to cover key links such as spatial offset calculation after place name alias ambiguity resolution, event quintuple extraction, and interactive trajectory rendering. In real-time scenarios such as multi-source public opinion and emergency dispatch, recall breakpoints, semantic-spatial disconnect, and visualization vacuum are prone to occur, becoming the main technical bottlenecks for industrial implementation. Summary of the Invention

[0008] To address the problems existing in the prior art, this invention provides a target localization and event tracking visualization system and method based on a large language model.

[0009] This invention is implemented as follows: a target localization and event tracking visualization system based on a large language model, comprising: The user input processing module receives unstructured natural language input from users, extracts user intent, and completes the scheduling and processing flow initialization of downstream modules. The RAG-based large language model event extraction module integrates the Retrieval-Augmented Generation (RAG) framework with the Large Language Model (LLM) to automatically identify and extract structured event information from natural language. The place name resolution and spatial positioning module standardizes and accurately maps geographic information entities in natural language to geographic coordinate system, providing support for subsequent event spatial positioning and visualization. The map service and event visualization module consists of two independent services, which are responsible for the static service of map tile layers and the visualization rendering of event trajectories, respectively; used for event location marking, trajectory visualization and interactive map display.

[0010] Furthermore, the user input processing module supports diverse user query expressions. After preliminary semantic parsing, the user input content transforms the user question into a standard retrieval expression or embedding vector. The system automatically calls the RAG-based semantic retrieval and event extraction module to perform context generation and structured analysis processing.

[0011] Furthermore, the RAG-based large language model event extraction module automatically identifies and extracts structured event information from natural language, including: (1) Semantic retrieval stage: Access local / remote corpora (such as news, social media, announcements, reports); Based on keyword retrieval (BM25 algorithm) and vector retrieval (BGE vector representation model) algorithms, document paragraphs related to user questions are filtered from the knowledge base; (2) Structured extraction stage: Large language models are used to understand and structurally transform the retrieved text. Automatically identify the five-tuple information of an event: event name (event_name), time (time), initial geographical location (spot), orientation (orientation), and distance (distance); Output a list of events in a uniform format.

[0012] Furthermore, the place name resolution and spatial positioning module standardizes and accurately maps geographic information entities in natural language to a geographic coordinate system, including: (1) Place name analysis: First, a structured global geographic name database is built and saved in JSON format; (2) Place name search and matching methods: Fuzzy string matching: The token_sort_ratio function from the fuzzywuzzy library is used for multiple alias matching, taking into account word order perturbation and spelling errors.

[0013] (3) Spatial offset calculation (calculation of the actual latitude and longitude of the event): After parsing the initial geographical location (spot), orientation, and distance of the event, the system uses a spherical geometry algorithm (Haversine) to convert them into target location coordinates. .

[0014] Furthermore, the spherical geometry algorithm (Haversine) is used to convert it into the target location coordinates. The specific method is as follows: Assuming the given starting point coordinates are The distance is (in meters), bearing angle is The target point is calculated using the Haversine forward formula, which is as follows: in, kilometers (Earth's radius), Expressed in radians.

[0015] Furthermore, the static service of map tile layers and the visualization rendering of event trajectories in the map service and event visualization module include: (1) Map tile service construction: The local tile service deployment uses a lightweight HTTP interface built with Flask to read local tile folders according to the XYZ directory structure (supporting .jpg and .png formats); Supports map styles: Different map styles can be switched according to the tile catalog, including administrative base map, satellite map, military map, etc.; Interface format: Conforms to the standard XYZ request protocol, for example: / {z} / {x} / {y}.png, compatible with front-end frameworks such as Leaflet and OpenLayers.

[0016] (2) Event visualization service: Event trajectory visualization (based on Folium + Leaflet) automatically renders markers for the original location and the calculated target location, displaying structured information such as event name, time, and location; Use Folium PolyLine to implement directional lines connecting the original location to the target location, supporting dynamic highlighting of the trajectory; Multiple event target points are automatically sorted by time and then their trajectory paths are drawn (AntPath dynamic line effect); This module ultimately generates an HTML map file containing all trajectory information.

[0017] Another objective of this invention is to provide a method for target localization and event tracking visualization based on a large language model, comprising the following steps: Step 1: The user input processing module receives unstructured natural language input from the user, extracts the user's intent, and completes the scheduling and processing flow initialization of downstream modules. Step 2: The RAG-based large language model event extraction module automatically identifies and extracts structured event information from natural language; Step 3: The place name resolution and spatial positioning module standardizes the geographic information entities in natural language and accurately maps them to the geographic coordinate system; Step four: The map service and event visualization module displays all trajectory information in the map tile layer system.

[0018] Another object of the present invention is to provide a computer device including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the target localization and event tracking visualization method based on a large language model.

[0019] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the target localization and event tracking visualization method based on a large language model.

[0020] Another objective of this invention is to provide an information data processing terminal, which includes the aforementioned target localization and event tracking visualization system based on a large language model.

[0021] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows: First, this invention proposes an event spatial perception and visualization method that combines Large Language Model (LLM), Retrieval Enhancement Generation (RAG), and Geographic Information System (GIS) capabilities. This aims to address the numerous limitations of traditional rule-based or structured data processing methods in natural language understanding, geographic feature identification, and spatial mapping. Compared to existing technologies, this invention offers the following significant technical advantages and beneficial effects: 1. Possesses powerful automatic extraction capabilities for unstructured text event elements. Leveraging the capabilities of Large Language Models (LLMs) in context modeling and semantic understanding, this invention enables the automatic extraction of event elements (such as event name, time, location, direction, and distance) from complex natural language. Combined with a Retrieval Augmentation (RAG) mechanism, it effectively alleviates the limitations of traditional rule-based or template-based methods in terms of universality and robustness, and improves the accuracy of structured extraction in diverse text scenarios.

[0022] 2. Achieve standardized parsing and coordinate mapping for natural language geographic descriptions. This invention proposes a place name parsing mechanism that integrates fuzzy string matching, contextual semantic analysis, and geographic knowledge base retrieval. It can accurately disambiguate fuzzy geographic descriptions such as abbreviations, aliases, and polysemous place names, and standardize them into unique latitude and longitude expressions in the global coordinate system, significantly enhancing the stability and universality of the system for geographic location identification.

[0023] 3. Supports spatial reasoning computation capabilities based on natural language direction and distance information. In response to the prevalent relative spatial descriptions in natural language (such as "approximately 200 kilometers in the southwest direction"), this invention introduces azimuth calculation and the spherical geodesic formula (Haversine algorithm). Based on known reference points, directions, and distance information, it accurately infers the location of the target event, realizing the automatic conversion of relative spatial descriptions to absolute geographic coordinates, thus overcoming the computational blind spots of traditional text spatial analysis.

[0024] 4. Provides high-precision spatial visualization and interactive capabilities for structured event information. This invention designs a Web GIS-based visualization rendering module that supports geolocation of structured event information, dynamic trajectory drawing, and user interaction (such as event clicks and detail pop-ups). This module establishes a data loop between structured text information and spatial layer rendering, enhancing the intuitiveness, practicality, and decision support capabilities of event spatial analysis.

[0025] Second, this invention proposes a target localization and event tracking visualization system based on a large language model. This system automatically extracts event elements from unstructured text, parses geographic information and maps spatial coordinates, and completes map annotation and dynamic display of events. The main technical points protected are as follows: Event element extraction: Based on a large language model combined with a retrieval augmented generation (RAG) mechanism, it automatically identifies and structurally extracts key semantic elements such as event name, time, location, direction, and distance from natural language text. It has contextual understanding capabilities and is suitable for complex sentence structures and implicit semantics.

[0026] Place name disambiguation and standardization: To address the problem of ambiguous expressions such as abbreviations, polysemy, and aliases in geographical names in natural language, this paper adopts string fuzzy matching, context semantic comparison, and geographical knowledge base retrieval technology to accurately parse place names and map them to unique global latitude and longitude coordinates.

[0027] Relative spatial location analysis: By introducing spherical geodesic algorithms (such as the Haversine formula) and a direction angle conversion mechanism, it supports the calculation of the absolute geographical location of an event from a "relative description" (such as "50 kilometers northeast of a certain place"), and realizes the automatic conversion of azimuth-distance information to spatial coordinates.

[0028] Map Visualization and Interaction: Construct visualization components based on Web GIS to support accurate rendering of structured event information on maps, including event location marking, trajectory display, visualization of spatial evolution, and interactive operations between users and the map, such as clicking to view details and filtering specific event types.

[0029] A unified geographic entity coordinate system and indexing mechanism: A global geographic entity standard library is built based on OpenStreetMap data, integrating names, aliases, multilingual expressions and a unified coordinate system, supporting fast place name resolution and location query, and providing efficient geographic entity support for the system.

[0030] Through the collaborative work of the above-mentioned technical modules, this invention outperforms existing technologies in terms of the accuracy of event information extraction, the precision of geographic information parsing, the visualization interactivity, and the overall level of automation. It is applicable to a variety of complex application scenarios such as military situation analysis, news event tracking, and public safety emergency response. Attached Figure Description

[0031] Figure 1 This is a structural diagram of the target localization and event tracking visualization system based on a large language model provided in this embodiment of the invention; Figure 2 This is a schematic diagram of a target localization and event tracking visualization system based on a large language model provided in an embodiment of the present invention; Figure 3 This is a flowchart of the target localization and event tracking visualization method based on a large language model provided in this embodiment of the invention; Figure 4 This is a visualization diagram of user questions and standardized output of a large model provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the code definition for fuzzy query and precise positioning provided in an embodiment of the present invention; Figure 6 This is a battlefield situation analysis and decision-making intelligent agent diagram provided in the embodiments of the present invention.

[0032] Figure 7 This is a structured event extraction diagram provided in an embodiment of the present invention.

[0033] Figure 8 This is a target positioning and trajectory tracking diagram provided in an embodiment of the present invention.

[0034] Figure 9 This is a multi-track tracking and display diagram provided in the embodiments of the present invention.

[0035] Figure 10 This is a report generation diagram provided in an embodiment of the present invention.

[0036] Figure 11 This is a target trajectory display diagram of multi-turn question-and-answer sessions provided in an embodiment of the present invention.

[0037] Figure 12 This is a click display diagram of a specific trajectory target event provided in an embodiment of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] In scenarios such as emergency command, public opinion analysis, and logistics scheduling, the front end typically faces challenges such as fragmented textual intelligence, lack of spatiotemporal annotation, and low efficiency of manual comparison. Traditional rule- or dictionary-driven extraction engines experience a sharp drop in recall rates for colloquial and cross-domain corpora, preventing key event nodes from entering the decision-making chain in a timely manner. Meanwhile, geographical names in reports often appear as aliases, mix Chinese and English, and even include dialect translations, making it difficult for subsequent GIS systems to achieve integrated rendering. The industry urgently needs an end-to-end technology that can rapidly precipitate unstructured descriptions into interactive geographic elements.

[0040] This system significantly mitigates information omissions caused by inconsistent criteria by introducing a "dual-channel" retrieval mechanism—namely, parallel BM25 exact matching and BGE semantic vector nearest neighbor indexing—into the retrieval enhancement generation framework. The vector index, deployed on a GPU-accelerated FAISS instance using an HNSW graph structure, maintains latency within the tens of milliseconds range. The retrieved paragraphs, along with the original query, form an extended context, which is input into a large language model with 70 bytes of parameters. During the distillation-fine-tuning stage, the model is injected with a dedicated event quintuple template, ensuring that the output JSON fragments naturally meet subsequent parsing requirements, eliminating the need for traditional rule correction steps.

[0041] To further improve extraction accuracy, the structured transformation submodule predicts field boundaries token-by-token using autoregressive decoding and employs a pointer-generator mechanism to avoid entity truncation. The model inference process runs asynchronously in a containerized microservice, isolating high-concurrency calls through a Redis-based lightweight message bus. Each batch of inference results triggers the geospatial resolution pipeline immediately upon reaching the scheduling layer, achieving second-level data insertion. This approach reduces I / O blocking while maintaining data consistency between upstream retrieval and downstream location.

[0042] At the place name resolution level, a self-built global place name Gazetteer covering over 2.5 million entries is used, with all records mapped to the WGS-84 coordinate system and accompanied by a prefix-position inverted index. The fuzzy matching engine quantifies string edit distance based on token, sort, and ratio, and applies multi-level filtering from prefecture-level city to county to township, significantly reducing ambiguity due to identical names. For complex descriptions such as "35 kilometers northeast of Qingdao Port," the module first converts the azimuth to radians, then uses the Haversine forward calculation formula to obtain the target latitude and longitude, ultimately controlling the error within 120 meters, allowing direct integration with PostGIS spatial queries.

[0043] The visualization uses a Leaflet + WebGL overlay rendering scheme. The tile server is hosted on the Flask API, supporting XYZ paths and multi-level caching; trajectory connections are dynamically highlighted using hardware acceleration via PolyLine and AntPath; the entire map is serialized into self-contained HTML, which can be embedded in existing command platform iframes or deployed offline to edge nodes. Event nodes are presented dynamically in progressive order along a timeline, allowing decision-makers to replay complex event chains and perform What-if simulations within seconds.

[0044] This invention compresses the four-stage process of "retrieval-extraction-location-visualization" into a millisecond-level pipeline, solving long-standing issues in the spatialization of text events, such as recall gaps, place name ambiguities, coordinate drift, and rendering performance bottlenecks. Its modular architecture facilitates seamless integration with Kafka streaming bus, Prometheus monitoring, and enterprise-level IAM, enabling rapid deployment in industries such as media monitoring, smart ports, and disaster early warning, providing high-confidence, low-latency event situational awareness capabilities for real-time scenarios.

[0045] like Figure 1 As shown, this embodiment of the invention provides a target localization and event tracking visualization system based on a large language model, comprising: The user input processing module receives unstructured natural language input from users, extracts user intent, and completes the scheduling and processing flow initialization of downstream modules. The RAG-based large language model event extraction module integrates the Retrieval-Augmented Generation (RAG) framework with the Large Language Model (LLM) to automatically identify and extract structured event information from natural language. The place name resolution and spatial positioning module standardizes and accurately maps geographic information entities in natural language to geographic coordinate system, providing support for subsequent event spatial positioning and visualization. The map service and event visualization module consists of two independent services, which are responsible for the static service of map tile layers and the visualization rendering of event trajectories, respectively; used for event location marking, trajectory visualization and interactive map display.

[0046] Furthermore, the user input processing module supports diverse user query expressions; After initial semantic parsing, the user input is transformed into a standard search query or embedding vector. The system then automatically calls the RAG-based semantic retrieval and event extraction module to perform context generation and structured analysis.

[0047] Furthermore, the RAG-based large language model event extraction module automatically identifies and extracts structured event information from natural language, including: (1) Semantic retrieval stage: Access local / remote corpora (such as news, social media, announcements, reports); Based on keyword retrieval (BM25 algorithm) and vector retrieval (BGE vector representation model) algorithms, document paragraphs related to user questions are filtered from the knowledge base; (2) Structured extraction stage: Large language models are used to understand and structurally transform the retrieved text. Automatically identify the five-tuple information of an event: event name (event_name), time (time), initial geographical location (spot), orientation (orientation), and distance (distance); Output a list of events in a uniform format: Furthermore, the place name resolution and spatial positioning module standardizes and accurately maps geographic information entities in natural language to a geographic coordinate system, including: (1) Place name analysis: First, a structured global geographic name database is built and saved in JSON format; each record contains the following fields: The explanations of each field are as follows: (2) Place name search and matching methods: Fuzzy string matching: The token_sort_ratio function from the fuzzywuzzy library is used for multiple alias matching, taking into account word order perturbation and spelling errors.

[0048] (3) Spatial offset calculation (calculation of the actual latitude and longitude of the event): After parsing the initial geographical location (spot), orientation, and distance of the event, the system uses a spherical geometry algorithm (Haversine) to convert them into target location coordinates. .

[0049] Furthermore, the spherical geometry algorithm (Haversine) is used to convert it into the target location coordinates. The specific method is as follows: Assuming the given starting point coordinates are The distance is (in meters), bearing angle is The target point is calculated using the Haversine forward formula, which is as follows: in, kilometers (Earth's radius), Expressed in radians.

[0050] Furthermore, the static service of map tile layers and the visualization rendering of event trajectories in the map service and event visualization module include: (1) Map tile service construction: The local tile service deployment uses a lightweight HTTP interface built with Flask to read local tile folders according to the XYZ directory structure (supporting .jpg and .png formats); Supports map styles: Different map styles can be switched according to the tile catalog, including administrative base map, satellite map, military map, etc.; Interface format: Conforms to the standard XYZ request protocol, for example: / {z} / {x} / {y}.png, compatible with front-end frameworks such as Leaflet and OpenLayers.

[0051] (2) Event visualization service: Event trajectory visualization (based on Folium + Leaflet) automatically renders markers for the original location and the calculated target location, displaying structured information such as event name, time, and location; Use Folium PolyLine to implement directional lines connecting the original location to the target location, supporting dynamic highlighting of the trajectory; Multiple event target points are automatically sorted by time and then their trajectory paths are drawn (AntPath dynamic line effect); This module ultimately generates an HTML map file containing all trajectory information.

[0052] This invention combines the information extraction advantages of large language models with the spatial analysis capabilities of GIS, making the construction of a system that supports automatic extraction, spatial positioning, and dynamic visualization of target events a necessary path. By introducing Retrieval-Augmented Generation (RAG) technology, the model's ability to acquire external knowledge can be improved, assisting in the extraction of key information such as event name, time, location, direction, and distance. Furthermore, with the help of a geographic database and spatial parsing module, fuzzy geographic descriptions can be converted into mappable coordinate information, enabling map labeling and dynamic tracking of events.

[0053] This invention solves the problem of automatically extracting event elements (such as event name, time, location, direction, distance, etc.) from natural language by combining Large Language Model (LLM) and Retrieval Enhanced Generation (RAG) mechanism, and outputs standardized structured data format, laying the foundation for subsequent spatial positioning and analysis.

[0054] This invention integrates technologies such as fuzzy string matching, contextual semantic matching, and geographic knowledge base retrieval to construct a place name parsing and matching module, which effectively solves the problems of place name disambiguation and standardization, and supports the accurate conversion of place names into unique latitude and longitude coordinates under the global coordinate system.

[0055] This invention introduces azimuth conversion and spherical geodesic algorithm (Haversine formula) to calculate the actual coordinates of an event based on the initial geographic point, direction (azimuth), and distance, thus solving the technical problem of converting relative position information to absolute spatial coordinates.

[0056] Based on Web GIS technology, this invention designs a map visualization module that supports structured event rendering, solving the coupling problem between structured information and map rendering, and realizing functions such as event marking, trajectory presentation, and user click interaction, significantly improving the visibility and operability of spatial event information.

[0057] Based on OpenStreetMap data, this invention constructs a global local geographic entity index library with a unified coordinate representation and naming system. It supports fast retrieval based on name and alias, solves the key problems of missing place name indexes and standardized expressions, and provides a reliable foundation for spatial positioning and event visualization of the entire system.

[0058] In summary, the event spatial perception and visualization system built based on LLM and RAG technologies and combined with GIS capabilities will provide intelligent and efficient technical support for scenarios such as military analysis, news tracking, and emergency response, and has significant application value and development prospects.

[0059] like Figure 3As shown, this embodiment of the invention provides a method for target localization and event tracking visualization based on a large language model, comprising the following steps: Step 1: The user input processing module receives unstructured natural language input from the user, extracts the user's intent, and completes the scheduling and processing flow initialization of downstream modules. Step 2: The RAG-based large language model event extraction module automatically identifies and extracts structured event information from natural language; Step 3: The place name resolution and spatial positioning module standardizes the geographic information entities in natural language and accurately maps them to the geographic coordinate system; Step four: The map service and event visualization module displays all trajectory information in the map tile layer system.

[0060] This example provides a complete workflow from user inquiries to standardized output of a large model, accurate geolocation calculation, and final front-end page display.

[0061] In this embodiment, the system first performs a semantic relevance retrieval based on the user's input question, calling a locally built knowledge base to return several paragraphs closely related to the user's query. These paragraphs are then used as context input to a large language model, which extracts structured event element information through natural language understanding. The structured information extraction process is as follows: Figure 4 As shown.

[0062] Based on the structured event information returned by the large language model, the system further refines the fields such as spot (initial geographic location), orientation, and distance. Specifically, the spot field is compared with place name entries in a pre-built geographic information database using fuzzy matching to locate the latitude and longitude of the place name in the standard coordinate system. Then, combining this standard geographic location, the direction angle (in radians) represented by the orientation field, and the spatial distance (in nautical miles) provided by the distance field, the system uses an improved Haversine spherical triangulation algorithm to calculate the actual geographic coordinates of the event. The calculation process is as follows: Figure 5 As shown.

[0063] The calculated actual geographical location, along with the original place name location, will be visualized in the constructed map tile layer system. This layer system supports multi-level zooming, and users can interact with the map, including zooming in, zooming out, dragging, and line analysis. The system supports responding to click events at any marker point, thereby displaying detailed information such as the event name, time, and fleet type associated with that location, achieving efficient and intuitive situational awareness and event tracking capabilities.

[0064] In summary, this system automatically identifies the five-tuple information (event name, time, initial geographical location, direction, and distance) of target events from natural language text. Combining the information extraction capabilities of a large language model with the spatial positioning and visualization capabilities of a Geographic Information System (GIS), it constructs an event spatial perception and dynamic display system for practical application scenarios. The system can accurately parse fuzzy geographical descriptions and relative location information in text, automatically calculate precise geographical coordinates, and also implement functions such as visual annotation of events on maps, trajectory display, and user interactive queries. This system can effectively support multi-source information fusion and spatial analysis tasks such as military situation analysis, emergency response, and public opinion dynamic tracking, demonstrating good intelligence, practicality, and promotional value.

[0065] The implementation of this invention in military applications is integrated in a workflow manner, covering the entire process from knowledge retrieval driven by intelligent question answering to structured event extraction, geographic target localization, trajectory extrapolation, map display and report generation, thereby enhancing situational awareness and decision support capabilities.

[0066] Figure 6 It showcases battlefield situation analysis and decision-making intelligence.

[0067] Figure 7 It demonstrates structured event extraction. Figure 8 It demonstrates target localization and trajectory tracking.

[0068] Figure 9 It demonstrates multi-track tracking and display. Figure 10 The report generation was demonstrated.

[0069] Figure 11 It displays the user's target trajectory across multiple rounds of question-and-answer sessions.

[0070] Figure 12 It displays clicks on specific trajectory target events.

[0071] This embodiment demonstrates the deep integration of the present invention with Large Language Model (LLM) and Retrieval Augmentation Generation (RAG) technologies. For a natural language question posed by the user, the system first performs a semantic similarity search in a pre-built domain knowledge base to obtain highly relevant contextual information. Subsequently, the user question and the search results are input into the Large Language Model to complete the structured extraction of event elements.

[0072] The extracted event data is input into a target location and event tracking visualization system. This system relies on a geographic database, first performing an initial fuzzy matching query of place names, and then using structured fields (such as time, direction, and distance) to achieve precise coordinate positioning. Based on this, the system automatically performs event deduplication and merging (based on spatial similarity) and time sorting to construct a clear spatiotemporal evolution chain.

[0073] Ultimately, the system renders event trajectories and tag information using HTML templates to achieve dynamic map display, and generates a complete analysis report by combining the extracted structured results with contextual content, thereby supporting battlefield situation understanding and intelligent decision-making simulation.

[0074] This architecture is highly modular and scalable, making it suitable for a variety of complex scenarios, such as military situational awareness, emergency response analysis, and cross-regional spatiotemporal event tracking.

[0075] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0076] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A large language model-based target positioning and event tracking visualization system, characterized in that, The system comprises: a user input processing module for receiving unstructured natural language input and performing semantic parsing, intent recognition and process scheduling; a large language model event extraction module based on retrieval enhancement generation for understanding relevant text segments obtained through semantic retrieval and outputting structured event lists containing event name, time, starting location, direction and distance; a place name resolution and spatial positioning module for standardizing geographic information entities, alias matching and target coordinate calculation based on spherical geometry algorithms; a map service and event visualization display module including a local tile service submodule and an event trajectory rendering submodule for providing map tile access, event labeling, trajectory connection and interactive display.

2. The system of claim 1, wherein the user input processing module supports free text, voice and multi-language queries, can convert parsing results into standard search expressions or vector embeddings, and automatically calls the event extraction module to perform context generation and structured analysis.

3. The system of claim 1, wherein the event extraction module comprises: a semantic retrieval unit that filters relevant document paragraphs from local or remote corpora based on the BM25 algorithm and the BGE vector retrieval model; a structured extraction unit that uses a large language model to reason about the retrieved paragraphs and output event five-tuples.

4. The system of claim 1, wherein the place name resolution and spatial positioning module uses the fuzzy string matching algorithm token_sort_ratio in alias matching and the Haversine forward algorithm to obtain target latitude and longitude from starting latitude and longitude, direction angle and spherical distance in coordinate calculation.

5. The system of claim 1, wherein the map service and event visualization display module: provides local tile services conforming to the XYZ protocol through Flask, supporting administrative, satellite and military map styles; uses Folium and Leaflet to render event starting and target point markers, PolyLine connections and AntPath dynamic trajectories, and generates interactive HTML files containing all trajectory information.

6. A large language model-based target positioning and event tracking visualization method, comprising the following steps: a. Receive user natural language input and complete semantic parsing, intent recognition and module scheduling; b. Retrieve and extract structured event information containing event name, time, starting location, direction and distance from relevant text segments through the retrieval enhancement generation framework; c. Perform standardization and alias matching on geographic information entities, and calculate target coordinates based on spherical geometry algorithms; d. Render event markers, trajectory connections and interactive displays on map tile layers.

7. A computer program product containing a plurality of instructions that, when executed by a processor, cause the processor to perform all the steps of the method of claim 6.

8. A non-transitory computer-readable storage medium having stored thereon the computer program product of claim 7.

9. An electronic device comprising a processor, a memory and a communication interface, the processor configured to load and execute the computer program product of claim 7, thereby implementing a target positioning and event tracking visualization function.

10. A method for place name resolution and spatial positioning, comprising: i. constructing a structured database containing global place name entries; ii. using a fuzzy string matching algorithm to match place names in natural language with aliases; iii. using a Haversine forward algorithm to obtain target coordinate positions based on resolved starting coordinates, direction angles and spatial distances.

Images