Interactive tactical analysis method for team sports based on large language model

Abstract

The invention discloses a team motion interactive tactical analysis method based on a large language model, which comprises the following steps: carrying out multi-modal alignment on a visually edited tactical board and a tactical description text, and extracting tactical knowledge from the aligned tactical image-text; customizing a scene based on the provided visual elements, extracting scene knowledge from the scene, and fusing the scene knowledge and the tactical knowledge to form domain knowledge; constructing a prompt project template for tactical analysis of the large language model; natural language input is carried out on the basis of the provided structured labels, input prompts are formed on the basis of domain knowledge and the input natural language in combination with a prompt engineering template, and the large language model carries out tactical analysis reasoning on the basis of the input prompts and visualizes tactical analysis results. According to the method, a user is allowed to completely express tactical intentions including complex spatio-temporal context information to a large language model in a visual mode, iterative reasoning can be carried out on a thinking chain, and a tactical analysis result easy to understand is output.

Description

Technical Field

[0001] The present invention belongs to the technical field of interactive data analysis, and in particular relates to a team sports interactive tactical analysis method based on a large language model. Background Art

[0002] As a data analysis and presentation tool, visualization technology has been widely used in the analysis of various sports. Through graphics, charts, and interactive interfaces, complex patterns and trends in data can be understood and interpreted more intuitively. Many specialized visualization systems have emerged to meet the data types and analysis requirements of different sports. In the field of basketball, most research focuses on designing visualization systems to support game analysis.

[0003] Although existing work has made progress in different fields, how to combine visualization techniques with large language models (LLMs) for basketball tactical analysis remains an unsolved challenge. Although simulation tools such as BasketballGAN and Basketball Flow simulate defensive strategies through user-drawn offensive tactics, they lack detailed visualization analysis and cannot fully explain complex game scenarios.

[0004] Big language models are a class of natural language processing models based on deep learning that can understand and generate human-like language text. By training on large-scale text data, big language models can perform well in multiple domain tasks, including language understanding, generation, and complex data analysis. They are able to combine various data types and process complex information and tasks. However, when processing complex tasks, it is difficult to accurately convey the user's intention using only natural language, so it is necessary to create efficient visualization interfaces and interactive methods to help users better understand, control, and improve the output of big language models. For example, ChartSpark proposed a visualization system that helps users obtain high-quality data charts by interactively optimizing prompt words. To ensure that big language models can complete specific tasks based on predefined knowledge, C2Idea incorporates design principles into prompt words, enabling big language models to create color schemes that follow these principles. To help users better understand the output of big language models and how they work, CommonSenseVis developed an interactive visualization system to explain and explore the common sense reasoning ability of natural language models. There are also some studies that treat big language models as big language models to support collaborative analysis and show that big language models can help process complex information and facilitate decision-making processes. However, its application in promoting sports analysis, especially basketball tactical analysis, has not been fully explored. Although large language models have strong capabilities in interpreting spatiotemporal data, it is still a challenge to enable experts to effectively explore and analyze tactics with the help of large language models. Moreover, its application in tactical design and simulation is still in the exploratory stage, and existing models have certain limitations in incorporating domain knowledge and the quality of simulation results.

[0005] In sports, tactics are usually composed of multiple complex components that are constantly changing during the game, making tactical analysis extremely complex. In order to evaluate the effectiveness of tactics in the game, some studies have combined professional domain knowledge to build models to improve tactical analysis. Traditional statistical models rely on quantitative data to evaluate tactical effects, but these models often have difficulty in fully capturing the complexity and dynamic changes of tactics. Although deep learning artificial intelligence models have shown some potential in tactical analysis, modeling the complex interactions between each player's action and the final tactical effect remains challenging.

[0006] In terms of tactical visualization and visual analysis, a series of studies have developed interactive visualization systems. For example, Tac-Simur designed a visual analysis system based on a second-order Markov chain to calculate tactical probabilities for simulation analysis of table tennis tactics. OBTracker proposed an interpretable off-ball movement model to evaluate individual contributions and team patterns in basketball off-ball tactics, and designed an interactive visual analysis system to support the interpretation and exploration of the results. Although these visualization methods can help users better understand the results of model analysis, users still need to perform complex logical reasoning to effectively align the actions of players at each step of the tactics with the final tactical effect for fine-grained analysis. Summary of the invention

[0007] In view of the above, the purpose of the present invention is to provide an interactive tactical analysis method for team sports based on a large language model, which allows users to fully express their tactical intentions to the large language model in a visual way, including complex spatiotemporal context information, and can perform iterative reasoning through a chain of thoughts to output easy-to-understand tactical analysis results.

[0008] To achieve the above-mentioned purpose of the invention, an embodiment provides a team sports interactive tactical analysis method based on a large language model, comprising the following steps:

[0009] Perform multimodal alignment on the visually edited tactical board and tactical description text, and extract tactical knowledge from the aligned tactical image and text pairs;

[0010] Customize scenarios based on the provided visualization elements and extract scenario knowledge from the scenarios, integrating scenario knowledge and tactical knowledge to form domain knowledge;

[0011] Construct a prompt engineering template for a large language model to perform tactical analysis, including specifying the role of the large language model, specifying the specific domain knowledge that the large language model should consider when analyzing, clarifying the tasks of the large language model, defining the output format for the large language model, providing multi-modal input for the large language model, and requiring the large language model to reason according to the thought chain prompts;

[0012] Natural language input is performed based on the provided structured tags, and input prompts are formed based on domain knowledge and the input natural language in combination with prompt engineering templates. The large language model performs tactical analysis reasoning based on the input prompts and visualizes the tactical analysis results.

[0013] Preferably, the visually edited tactical board contains tactical visual elements, specifically including player symbols, trajectory lines, and action arrows, and extracts visual information through time and space for multimodal alignment, specifically records the mapping of visual information relationships on the tactical board through a coordinate system, locates the position of the player on the court through a coordinate system, and uses continuous coordinate points to represent the trajectory, and at the same time describes the order of actions according to the order edited on the tactical board to determine the logic of tactical execution.

[0014] Preferably, the provided visualization elements include text, charts, and tables, the scene is customized based on these visualization elements, and the scene knowledge extracted from the scene includes physical fitness information, statistical information, and behavioral information;

[0015] Among them, physical fitness information includes the height, weight data, and average speed of each player in the current scene; statistical information is used for the interactive selection of team and player matchups in the current tactics, and on-court / off-court statistics are calculated for each selected player matchup, including five indicators: field goal hits, three-pointer hits, assists, turnovers, and personal fouls; behavioral information is reflected in the scene, allowing users to retrieve similar scenes and link similar tactical executions in real games with the current scene.

[0016] Preferably, the method further comprises: the domain knowledge generates an embedding vector through an embedding model and stores the embedding vector in a vector library for retrieval by a large language model, wherein the embedding model comprises a text-embedding-3-large model.

[0017] Preferably, the tasks of the large language model include recommendation tasks, explanation tasks, and evaluation tasks, and when defining the output format for the large language model, common variables and task-specific variables are defined for the three categories, wherein the common variables include the tactical role and starting position of the player, and the task-specific variables include the event type for the recommendation task, the cause and goal of the event for the explanation task, and the event result and countermeasure for the evaluation task.

[0018] Preferably, when the large language model performs reasoning according to the thought chain prompts, it performs four levels of reasoning, including context-based environment level reasoning, team level reasoning, player level reasoning, and action-based behavior level reasoning.

[0019] Preferably, it also includes: in the tactical analysis reasoning process, the tactical analysis results for each input prompt are saved as historical nodes and visualized, and used as part of the large language model input in subsequent interactions, to ensure that the large language model uses instance information to generate coherent and relevant tactical analysis results.

[0020] Preferably, the structured tags provided include description tags, requirement tags, and comment tags. The description text input based on the description tags is used to supplement the input, and the description text is added to the input component of the prompt. The requirement text input based on the requirement tags is integrated into the task component. The comment text input based on the comment tags is used as feedback on the tactical analysis results and is input together with the current tactical analysis results for the next tactical analysis reasoning.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] Customize scenarios and tactics based on visual tactical boards and elements and build domain knowledge. Input tactical-related information to be analyzed based on constructed prompt engineering templates and structured tags. Perform multi-level and multi-task tactical analysis and reasoning through a large language model, and visualize the reasoning and analysis results to explore, recommend, and explain tactical designs. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A basketball tactic analysis framework that supports complex interactions with a large language model is provided in an embodiment;

[0025] Figure 2 is a flow chart of a method for interactive tactical analysis of team sports based on a large language model provided by an embodiment;

[0026] Figure 3 is a multimodal alignment flow chart provided by the embodiment, which includes two multimodal alignment templates. We use colors to describe the information after visual extraction: orange represents the player character, blue represents the action, green represents the position, and red represents the trajectory;

[0027] Figure 4 is a knowledge integration flow chart provided by an embodiment, wherein A represents tactic-based knowledge and B represents scenario-based knowledge;

[0028] Figure 5 is a schematic diagram of a tactical analysis prompt engineering template provided in an embodiment;

[0029] Figure 6It is an interactive interface for interactive tactical analysis of team sports provided by an embodiment, wherein A is a chat view that provides system feedback through tag selection and open-ended questions and answers to enhance communication between users and the system; B is a deployment view that provides interaction during the tactical setting process, including tactical drawing, position analysis, and scenario retrieval; C is a simulation view that displays the tactics recommended by the intelligent coach and provides an overview and detailed explanation and evaluation; D is a history view that records the user's tactics and provides classic tactics as a starting point for exploration. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the protection scope of the present invention.

[0031] The embodiment of the present invention proposes to use a large language model and visualization technology to solve the problem in the background technology. The large language model has demonstrated strong capabilities in understanding complex text descriptions and processing spatiotemporal data. This shows that such large language models can not only understand complex tactical text descriptions, but also provide forward-looking insights through tactical reasoning, surpassing the capabilities of traditional models. In this embodiment, basketball, a representative sport, is selected as an application scenario. In collaboration with basketball experts, a method for interactive tactical analysis of team sports based on a large language model is proposed. Users can draw on a tactical board to define the tactics of interest. Through visualization, users can further specify tactical scenarios, such as score differences or specific players and teams. An intelligent coach based on a large language model will receive these multimodal inputs and use its basketball tactical knowledge to simulate the situation on the court. The output of the intelligent coach will be converted into a visual form, and the user can gradually interact with the large language model to understand the tactical formulation process and its possible variations. The visual analysis method significantly reduces the level of knowledge required by the user because the intelligent coach can help the user perform complex tactical design and analysis tasks.

[0032] However, there are also many challenges in directly using large language models for interactive tactical analysis. During the development process, two major challenges were encountered. The first challenge is to propose a method for fine-grained tactical analysis using large language models. In the case of diversity in the tactical understanding of basketball experts, customized input requirements are required. In addition, in order to enhance the realism of the simulation environment, the results should present coherent and easy-to-understand decision steps. However, the limited domain knowledge of large language models seriously affects the quality of simulation results, which makes this task challenging. The second challenge is to provide effective exploration, recommendation, and explanation functions to support tactical design. Experts want to explore simulation scenarios as comprehensively as possible, but it is very difficult to foresee all possible scenarios during the simulation. Therefore, recommendations need to be provided based on the current context. In addition, since the output form of large language models is limited to natural language, it is difficult for users to quickly extract key information from a large amount of text output. Therefore, it is particularly important to convert text information into visualizations and provide explanations. It is very challenging to design an interactive visualization system based on large language models to support exploration, recommendation, and explanation of tactical design.

[0033] Based on this, the embodiment provides a team sports interactive tactical analysis framework based on a large language model, such as Figure 1 As shown, it includes user multimodal input, model intelligent reasoning and iterative exploration process.

[0034] In user multimodal input, background information provides necessary support for detailed tactical analysis. In order to help users fully convey information and intentions to the large language model, hand-drawn tactical board sketches are introduced to further promote information integration. This creates a flexible interactive environment for basketball tactical design. Visualization enhances user input from a data perspective. Text, charts, and tables are the basic forms of visualization, which are able to represent various types of data and make it easier for users to understand. Users can customize tactical analysis scenarios by interacting with visualization elements. Visualization provides solid support for data by integrating a wider range of background information into tactical settings.

[0035] In model intelligent reasoning, demand-driven output is crucial in the tactical exploration process. It enables the large language model to reason step by step to understand the complex factors provided by the user. In addition, the task categories of the large language model are summarized according to various needs. Reasoning requires the large language model to conduct four levels of top-down analysis: first, the background layer, focusing on the background of the game and specific scenes; second, the team layer, considering the overall strategy and style of the team; next, the player layer, evaluating the technical ability and performance of individual players; and finally, the action layer, focusing on the details of tactical actions and their execution. The step-by-step reasoning of chain thinking enables the large language model to consider multiple factors and perspectives of complex tactics. Task decomposition helps the large language model to break down complex tasks into more focused components. Tasks are decomposed into three categories: recommendation tasks, explanation tasks, and evaluation tasks. Recommendation tasks enable the model to provide various options and variations to provide tailored suggestions for tactical settings. Reasoning interpretation explanation tasks provide insights into the decision-making process of the large language model, significantly enhancing the transparency and understanding of tactics. Evaluation tasks involve a comprehensive analysis of simulation results to help users evaluate the potential and applicability of each setting.

[0036] In the iterative exploration process, the framework of the present invention supports complex iterative interactions in tactical exploration. User input serves as tactical settings, guiding the model towards customized scenarios. Model output provides insights and alternatives for tactical simulation, effectively transforming the analysis process into a process that cycles between settings and simulations, representing the reasoning evolution of tactics. Through this structured and flexible approach, users and large language models explore, simulate, and explain tactics in a collaborative way, promoting the understanding and optimization of tactics.

[0037] Based on the proposed interactive framework, an interactive visual analysis method and system Smartboard based on a large language model was also developed. This method and system can support users to personalize tactical analysis scenarios and conduct iterative exploration in the processes of tactical deployment, tactical simulation, and tactical evolution.

[0038] like Figure 2 As shown, the method for interactive tactical analysis of team sports based on a large language model provided by the embodiment includes the following steps:

[0039] S1, perform multimodal alignment on the visually edited tactical board and tactical description text, and extract tactical knowledge from the aligned tactical image and text pairs.

[0040] Multimodal alignment can effectively enhance the visual understanding ability of large language models by forming image-text pairs. In order to align tactical images and scene texts from different modalities, tactical visual elements are extracted from the visually edited tactical board sketch and combined with semantic descriptions to achieve effective multimodal alignment, such as Figure 3As shown. The tactical board contains tactical visual elements, including player symbols, trajectory lines, and action arrows. In order to extract the visual information in the tactical board in detail, the visual information is extracted through time and space for multimodal alignment, and the mapping of the visual information relationship on the tactical board is recorded through the coordinate system. Specifically, the position of the player on the court is located through the coordinate system, and the trajectory is represented by continuous coordinate points to better express the spatial relationship. At the same time, the action sequence is described according to the order of the tactical board sketch, that is, the order of editing on the tactical board to describe the logic of tactical execution.

[0041] In the specific multimodal alignment, two sets of templates are used to organize the extracted tactical visual elements corresponding to the player's movement trajectory and position respectively for semantic integration with the tactical description text. In order to enhance the comprehensibility of the overall description, the template includes the interaction between players and integrates semantic information to complete the alignment.

[0042] S2, customizes the scene based on the provided visualization elements and extracts scene knowledge from the scene, integrating scene knowledge and tactical knowledge to form domain knowledge.

[0043] Domain knowledge is a key factor affecting the performance of large language models. Although large language models perform well in some domain-specific tasks, they may lack sufficient expertise in tactical analysis of specific scenarios. To address this limitation, tactical knowledge and scenario knowledge are converted into knowledge documents, and the retrieval-augmented generation (RAG) method is used to enhance the responsiveness of large language models by retrieving relevant information, such as Figure 4 shown.

[0044] The constructed domain knowledge document consists of tactical-based knowledge (i.e., tactical knowledge) and scenario-based knowledge (i.e., scenario knowledge), organized in the form of text data. This provides comprehensive information support for the understanding and response generation of the large language model in tactical scenarios.

[0045] like Figure 4 The tactical knowledge shown in A in the figure is derived from the tactical image-text pair after multimodal alignment, including visual description and coordinate system. The visual description includes the visual encoding of the tactical board, such as symbols, lines, and arrows. The coordinate system includes the mapping of data relationships on the tactical board. This knowledge helps the large language model understand tactical details.

[0046] like Figure 4The scenario-based knowledge shown in B is designed to allow users to convey customized scenarios to the large language model. This type of knowledge covers various aspects of tactical scenarios, including physical information, statistical information, and behavioral information to provide a comprehensive understanding of tactical scenarios. The details are as follows: Physical information includes height and weight data for each player in the current scenario. In addition, the average speed of each player is calculated through the selected scenario and used as physical information to evaluate the physical performance of the player in the tactic. Statistical information is used for interactive selection of teams and player matchups in the current tactic. Specifically, on-court / off-court statistics are calculated for each selected player matchup. On-court / off-court information measures the difference in performance between offensive and defensive players on or off the court, helping users understand the impact of matchups on tactics. Five indicators, including field goals, three-pointers, assists, turnovers, and personal fouls, are also calculated and normalized. Behavioral information is reflected in the scenario, allowing users to perform similar context retrieval, which allows users to associate similar tactical executions in real games with the current scenario. Behavioral information helps users customize tactical scenarios so that the large language model can understand and generate similar scenarios.

[0047] After building the domain knowledge, enhanced generation is also performed. Specifically, embedding models such as the text-embedding-3-large model can be used to embed domain knowledge documents. The generated embedding vectors are stored in the FAISS vector library, allowing large language models to efficiently retrieve these vectors. This method enhances the ability of large language models to access and integrate relevant external knowledge, thereby improving the quality of their responses.

[0048] S3, a prompt engineering template for building a large language model for tactical analysis, including specifying the role of the large language model, specifying the specific domain knowledge that the large language model should consider during analysis, clarifying the tasks of the large language model, defining the output format for the large language model, providing multi-modal input for the large language model, and requiring the large language model to reason according to the thought chain prompts.

[0049] Hint engineering has been shown to be effective in improving the performance of large language models in solving domain-specific tasks. Figure 5 The dynamic prompt engineering template shown is used to generate content based on user needs using a large language model. The template includes six parts:

[0050] (1) Specify the role of the large language model. The role involves a context description related to the large language model, specifically and clearly defines the role played by the large language model, and specifies the capabilities required for the role.

[0051] (2) specifying specific domain knowledge that the large language model should consider during analysis, which enhances the response of the large language model by allowing the large language model to retrieve specific domain knowledge from the provided external knowledge documents;

[0052] (3) Clarify the tasks of the large language model, which include the requirements of each task. For example, the recommendation task requires recommending tactical settings, the explanation task deeply explores the logic behind tactical decisions, and the evaluation task focuses on analyzing the capabilities required to execute tactics and possible results;

[0053] (4) Define the output format for the large language model. The output format describes the specific output of each task. Specifically, a set of common variables and task-specific variables are defined for all three types of tasks. For example, the recommendation task requires an additional output action type. Both the explanation task and the evaluation task require an overview and detailed output. The overview considers the overall tactics, while the details focus on individual players. For the explanation, the large language model describes each player's actions in detail, including how these actions counter the tactics and their goals. The evaluation details explore whether the player's ability is consistent with the goal of the current action and evaluates potential opponent reactions.

[0054] (5) Provide multimodal input to the large language model. Input is the trigger for communication with the large language model. Basic input requires knowledge and image-text pairs. Considering the relevance of the tasks, the explanation task requires external input for recommendation output, while the evaluation task requires input for recommendation and explanation.

[0055] (6) The large language model is required to reason according to the chain of thought prompts. Reasoning is used to guide the logical reasoning process. Chain thought prompts allow the large language model to think and solve problems comprehensively. Specifically, the CoT method is used to guide the large language model to reason along a specific logical path, and the model is required to reason at four levels, including the context-based environment level, team level, player level, and action level.

[0056] S4, based on the provided structured tags, performs natural language input, forms input prompts based on domain knowledge and the input natural language and combines them with prompt engineering templates, and the large language model performs tactical analysis reasoning based on the input prompts and visualizes the tactical analysis results.

[0057] In order to flexibly control the generation of large language models, structured tags are provided to embed the user's natural language input. These tags are divided into three aspects according to different goals, namely description tags, requirement tags, and comment tags.

[0058] For descriptive tags, the user's description is mainly used to supplement the input. It is added to the input component of the prompt to enrich the large language model's understanding of the current situation.

[0059] For demand-type labels, user requirements are integrated into task components, which allows users to add semantic requirements for expected responses to ensure that the output of the large language model is closer to the user's specific requirements.

[0060] For the comment class label, the user's comments are fed back as the tactical analysis results of the response. The comments are added together with the current response tactical analysis results to guide the large language model to improve its response after integrating these comments. Considering the previous context, the large language model will refer to these comments when completing the task.

[0061] Based on domain knowledge and natural language input based on structured tags and combined with prompt engineering templates to form input prompts, the large language model performs tactical analysis reasoning based on the input prompts and visualizes the tactical analysis results. During the tactical analysis reasoning process, the tactical analysis results for each input prompt are saved as historical nodes and visualized, and used as part of the large language model input in subsequent interactions, ensuring that the large language model uses instance information to generate coherent and relevant tactical analysis results.

[0062] The above-mentioned interactive tactical analysis method for team sports based on large language models uses Figure 6 The system Smartboard shown in the figure consists of four views, where A is a chat view, through A1-A3, users can enter text to control the iterative exploration process with the large language model, B is a deployment view, which helps users define the initial tactics of interest, through B1 users can hand-draw the deployment details of the tactics, through B2 and B3 users can select the match information and view similar historical game situations. C is a simulation view, which displays the simulation alternatives returned by the large language model. C1 represents the visual content of each alternative, C2 represents the specific explanation and evaluation information, and the user can select an alternative to iterate the tactical simulation. D is a history view that records the simulation path D1 and the user's interaction history D2.

[0063] The embodiment verifies the technical effect of the present invention through case studies and user experiments based on actual basketball data sets, and demonstrates the usability of the present invention in analyzing team tactics.

[0064] Specifically, two experts with many years of coaching experience were invited to conduct case studies using the system of the present invention. These studies were based on data from the NBA 2015-16 season. After introducing the interactive and visual features of Smartboard, the experts freely explored the tactics and scenarios of interest for an hour, recorded their operations and comments, and collected their feedback through interviews to further improve the system. The first expert conducted an in-depth analysis of the "Bullhorn Tactics" and explored how to use screens and cuts to create scoring opportunities through tactical deployment and simulation. The expert found that creating open space through team movement was particularly important for the Detroit Pistons and adjusted the tactics according to different defensive strategies. The expert also explored the "Flex" tactic and found that it optimized the scoring opportunities of the Detroit Pistons players and verified its effectiveness through actual game data. The second expert studied the "Delay" tactic by simulating defensive strategies against the powerful offensive team Houston Rockets. The expert found that the high pick-and-roll against Howard required special defensive adjustments and proposed the principles of the delay tactic to effectively deal with the problem of misalignment in defense.

[0065] Experts gave positive comments on the overall effect of Smartboard, believing that it is very effective in basketball tactical analysis, providing in-depth tactical analysis and more exploration of complex tactical changes.

[0066] A user study was also conducted, inviting 8 experts in the field of basketball tactics to evaluate the tactical analysis content in Smartboard. Six indicators including accuracy, effectiveness, diversity, practicality, insight and consistency were established, and the scores of all participants on the indicators were collected. The score analysis results showed that all participants agreed that the tactical analysis content generated by Smartboard performed well in these indicators, confirming the ability of the invention in these indicators.

[0067] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for interactive tactical analysis of team sports based on a large language model, characterized in that: The following steps are involved: Perform multimodal alignment on the visually edited tactical board and tactical description text, and extract tactical knowledge from the aligned tactical image and text pairs; Customize scenarios based on the provided visualization elements and extract scenario knowledge from the scenarios, integrating scenario knowledge and tactical knowledge to form domain knowledge; Construct a prompt engineering template for a large language model to perform tactical analysis, including specifying the role of the large language model, specifying the specific domain knowledge that the large language model should consider when analyzing, clarifying the tasks of the large language model, defining the output format for the large language model, providing multi-modal input for the large language model, and requiring the large language model to reason according to the thought chain prompts; Natural language input is performed based on the provided structured tags, and input prompts are formed based on domain knowledge and the input natural language in combination with prompt engineering templates. The large language model performs tactical analysis reasoning based on the input prompts and visualizes the tactical analysis results.

2. The method for interactive tactical analysis of team sports based on a large language model according to claim 1, characterized in that: The visually edited tactical board contains tactical visual elements, including player symbols, trajectory lines, and action arrows, and extracts visual information through time and space for multimodal alignment, and records the mapping of visual information relationships on the tactical board through a coordinate system. Specifically, the player's position on the court is located through a coordinate system, and the trajectory is represented by continuous coordinate points. At the same time, the action sequence is described according to the order edited on the tactical board to determine the logic of tactical execution.

3. The method for interactive tactical analysis of team sports based on a large language model according to claim 1, characterized in that: The provided visualization elements include text, charts, and tables. The scene is customized based on these visualization elements, and the scene knowledge extracted from the scene includes physical fitness information, statistical information, and behavioral information; Among them, physical fitness information includes the height, weight data, and average speed of each player in the current scene; statistical information is used for the interactive selection of team and player matchups in the current tactics, and on-court / off-court statistics are calculated for each selected player matchup, including five indicators: field goal hits, three-pointer hits, assists, turnovers, and personal fouls; behavioral information is reflected in the scene, allowing users to retrieve similar scenes and link similar tactical executions in real games with the current scene.

4. The method for interactive tactical analysis of team sports based on a large language model according to claim 1, characterized in that: Also includes: The domain knowledge generates an embedding vector through an embedding model and is stored in a vector library for retrieval by a large language model, wherein the embedding model includes a text-embedding-3-large model.

5. The method for interactive tactical analysis of team sports based on a large language model according to claim 1, characterized in that: The tasks of the large language model include recommendation tasks, explanation tasks, and evaluation tasks. When defining the output format for the large language model, common variables and task-specific variables are defined for the three types of tasks. Common variables include the tactical role and starting position of the players, and task-specific variables include the event type of the recommendation task, the cause and goal of the event of the explanation task, and the event result and countermeasures of the evaluation task.

6. The method for interactive tactical analysis of team sports based on a large language model according to claim 1, characterized in that: When the large language model performs reasoning according to the thought chain prompts, it performs four levels of reasoning, including environmental level reasoning based on context, team level reasoning, player level reasoning, and behavioral level reasoning based on action.

7. The method for interactive tactical analysis of team sports based on a large language model according to claim 1, characterized in that: Also includes: During the tactical analysis reasoning process, the tactical analysis results for each input prompt are saved as history nodes and visualized, and used as part of the large language model input in subsequent interactions, ensuring that the large language model uses instance information to generate coherent and relevant tactical analysis results.

8. The method for interactive tactical analysis of team sports based on a large language model according to claim 1, characterized in that: The structured tags provided include description tags, requirement tags, and comment tags. The description text input based on the description tag is used to supplement the input and is added to the input component of the prompt. The requirement text input based on the requirement tag is integrated into the task component. The comment text input based on the comment tag is used as feedback on the tactical analysis results and is input together with the current tactical analysis results for the next tactical analysis reasoning.

Images