Basketball player evaluation and analysis method and system, electronic equipment and storage medium

Abstract

The invention discloses a basketball player evaluation and analysis method and system, electronic equipment and a storage medium. The method comprises the steps of data acquisition and acquisition; competition technical statistical items are collected; preprocessing the data; constructing a database, and storing the data into a database; an LLM-Basketball model is called for analysis, and a final answer is given; and the front end performs question and answer and chart display. According to the method, multi-dimensional data such as physiology, athletic ability and the like are introduced to be combined with the LLM-Basketball model, so that long-term tracking of performance or cross-season ability evaluation of the athletes is realized, and effective suggestions and decision support are provided for the athletes.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology for basketball sports evaluation, and in particular to a method, system, electronic device, and storage medium for evaluating and analyzing basketball players. Background Technology

[0002] Currently, in the field of sports competition, evaluation and analysis mainly focus on match prediction and the performance of participating teams or individuals, relying primarily on historical data of teams and players, match videos, and other data based on algorithms or models for evaluation. Invention Announcement - CN120375253 A - A real-time AI evaluation system for tennis based on multi-view visual feature fusion utilizes a temporal graph convolutional network to fuse multi-view video stream data of a match to generate spatiotemporal feature representations. Based on the fused spatiotemporal features, a temporal causal graph model is constructed. Graph structure analysis and event influence assessment are used to identify key turning points in the match, and personalized tactical suggestions are generated by combining player feature models. Invention Announcement - CN 120375253A - A method, device, and electronic device for evaluating and predicting the situation of a badminton singles match. This method processes the raw data from the first half of the game, the most recent pre-match physical fitness test data, and match results from the past year to extract secondary indicator features and potential distribution features, constructing a match situation evaluation and prediction model for a comprehensive assessment of the match outcome.

[0003] With the continuous development of artificial intelligence technology, Large Language Models (LLMs) are widely used in various fields, possessing capabilities such as key information extraction, data processing, and autonomous analysis, and can provide decision-making basis for complex problems through logical reasoning. Some research has attempted to use LLMs for sports match analysis. For example, in the invention publication CN119357590A – "Match Evaluation Processing Method and Device," player field position data and individual player data are input into the LLM to obtain player cooperative defensive ability characteristics, and then combined with player relationship characteristics, team characteristics, etc., to construct a match evaluation model for assessment. However, this method is mainly based on instantaneous data from a single match, lacking tracking of players' long-term performance, physical changes, stability, and other dimensions, and cannot provide long-term player performance analysis or cross-season ability assessment.

[0004] Therefore, there is an urgent need for an evaluation method that can integrate multi-dimensional features to analyze both single-game performance and track long-term development. To address these issues, this invention creates an LLM-Basketball model that incorporates physiological, physical, and body indicator data for long-term tracking, providing effective suggestions and guidance for basketball monitoring and game evaluation. Summary of the Invention

[0005] To address the shortcomings mentioned above, this invention provides a method, system, electronic device, and storage medium for evaluating and analyzing basketball players.

[0006] To achieve the above objectives, the present invention provides a method for evaluating and analyzing basketball players, comprising:

[0007] Collect athletes' exercise data, including physiological data, athletic ability data, physical indicators data, and recovery data;

[0008] Collect statistical data on game techniques, including the basic rules and scoring methods of basketball;

[0009] The motion data and statistical data are preprocessed and converted to obtain data in the required format.

[0010] The converted data is stored in the database based on the data source and type.

[0011] Based on the data in the database, the LLM-Basketball model is used to parse and generate workflow diagrams, node prompts, and operation codes.

[0012] Executing the aforementioned operation code yields the final answer;

[0013] The front end displays questions and answers as well as charts.

[0014] Preferably, the physiological data includes heart rate, TRIP, and calories; the exercise capacity data includes training load, training intensity, number of jumps, and number of changes of direction; the body indicator data includes age, height, weight, and BMI; and the recovery data includes recovery score and fatigue level.

[0015] Preferably, the statistical data includes 2 points scored, 3 points scored, free throws, fouls, turnovers, steals, and assists.

[0016] Preferably, the preprocessing of the motion data and the statistical data includes data cleaning, data calculation, and format conversion;

[0017] Data cleaning refers to the process of processing and organizing data, including consistency checks, handling missing values, and removing duplicate data.

[0018] The data calculation refers to converting the collected raw data into other data through calculation formulas;

[0019] The format conversion refers to converting the data into the format required for subsequent analysis.

[0020] Preferably, the process of generating workflow diagrams, node prompts, and operation codes by calling the LLM-Basketball model based on the data in the database includes:

[0021] Input a basketball-related question into the LLM-Basketball model;

[0022] Based on the instantiated LLM-Basketball model, a prompt parameter is constructed, which consists of role, task, data location, and specific requirements for constructing the graph.

[0023] The LLM-Basketball model's reasoning capabilities are invoked to analyze and generate thought processes;

[0024] Execute the Python code to generate the mind chain, which is used to generate the workflow graph (work_graph);

[0025] Execute the code to generate a work graph;

[0026] Execute the work_graph code to generate a directed acyclic graph in GraphML format to represent the workflow;

[0027] The node prompts are generated from a structured template and consist of three parts: description, function definition, and return line. The description is obtained from the node description in the workgraph, the function definition is obtained based on the predecessor node, and the return line is obtained based on the successor node.

[0028] Based on the prompt for each operation, the corresponding operation function is obtained. The operation function includes score statistics, player stability analysis, ranking statistics, shooting percentage prediction, and win rate prediction.

[0029] Send the requirements of each operation node to the LLM-Basketball model, request code implementation, and generate node instructions;

[0030] Based on the code template, the helper.py function of the LLM-Basketball model generates operation code.

[0031] Preferably, executing the operation code to obtain the final answer includes:

[0032] The execution operation is divided into 5 modules: user, Step model, Re-Step model, Operation, and Operation Link. The Step model is responsible for generating the specific workflow, the Operation model is responsible for executing each operation in the workflow, and the Operation Link model is responsible for connecting the entire workflow. During operation, the workflow needs to be dynamically adjusted based on the execution results of each step. The Re-Step model is responsible for modifying the workflow.

[0033] The workflow diagram is submitted to the Operation, and the operations are executed sequentially according to the steps. The execution of the operations is divided into two stages: the execution stage and the Re-Step stage.

[0034] After executing the first step of the workflow, the execution result of the first step is returned and placed into the historical execution results; the Operation Link submits the user issue, workflow, and historical execution results to the Re-Step model to generate a new workflow; after receiving the new workflow, the Operation Link executes the next operation;

[0035] Repeat the above execution and Re-Step steps until all work steps are completed;

[0036] Upon completion of the task, a final answer is returned, which may include questions and answers, tables, or diagrams.

[0037] The Operation Link displays the final answer to the user.

[0038] Preferably, the front-end includes Q&A and chart display, including:

[0039] The save_solution of the LLM-Basketball model is used to save the workflow as GraphML, and a REST endpoint is connected via FastAPI;

[0040] The results are transmitted to the backend via API, and the API is encapsulated.

[0041] The front-end calls the back-end interface and uses WebSocket for real-time interaction to achieve data stream transmission;

[0042] After the front end obtains the data stream, it renders it using the echarts chart component library to achieve question-and-answer and chart display.

[0043] This application also provides a basketball player evaluation and analysis system, including:

[0044] The first data acquisition module is used to collect athletes' exercise data, including physiological data, athletic ability data, physical indicator data, and recovery data.

[0045] The second data acquisition module is used to collect game statistics, including the basic rules and scoring methods of basketball games.

[0046] The preprocessing module is used to preprocess the motion data and the statistical data, and convert them to obtain data in the required format.

[0047] The storage module is used to store the converted data into a database based on the data source and type;

[0048] The parsing module is used to call the LLM-Basketball model based on the data in the database to generate workflow diagrams, node prompts, and operation codes;

[0049] The execution module is used to execute the operation code to obtain the final answer;

[0050] The presentation module is used for front-end Q&A and chart display.

[0051] The present invention also provides an electronic device, including at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the program is executed by the processing unit, the processing unit performs the method described above.

[0052] The present invention also provides a storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the above-described method.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] This invention combines multidimensional data such as physiological and athletic abilities with the LLM-Basketball model to achieve long-term tracking of athletes' performance or cross-seasonal ability assessment, providing athletes with effective suggestions and decision support. Attached Figure Description

[0055] Figure 1 This is a flowchart of the basketball player evaluation and analysis method of the present invention;

[0056] Figure 2 This is the LLM-Basketball-Plan Step flowchart of the present invention;

[0057] Figure 3 This is an example of the workflow of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0059] Reference Figure 1 This invention provides a method for evaluating and analyzing basketball players, comprising:

[0060] The LLM model can use ChatGPT4, Qwen3, or deepseek models; this embodiment uses the ChatGPT4 model. Specifically, it includes the following steps:

[0061] Step 1, Data Collection and Acquisition.

[0062] The collected data is divided into physiological data, athletic ability data, physical indicator data, and recovery data.

[0063] The physiological data includes parameters such as heart rate, TRMP, and calories, which are collected through wearable devices such as heart rate belts and smart bracelets.

[0064] The athletic performance data includes parameters such as training load, training intensity, number of jumps, and number of changes of direction, which are collected through devices such as motion tracking vests.

[0065] Body metrics data mainly include parameters such as age, height, weight, and BMI.

[0066] Recovery data includes parameters such as recovery score and fatigue level, which are collected through devices such as electromyography sensors.

[0067] Step 2: Collection of competition technical statistics.

[0068] Technical statistics are collected manually, including 2 points, 3 points, free throws, fouls, turnovers, steals, front rebounds, back rebounds, assists, and blocks.

[0069] Step 3: Preprocess the data.

[0070] The acquired physiological, athletic, physical, recovery, and technical statistics data were cleaned and transformed using Excel software.

[0071] Step 4: Build the database and store the data in it.

[0072] Build a basketball database and store the converted data into the database.

[0073] Step 5, call the LLM-Basketball model for analysis and decision-making, such as... Figure 2 As shown.

[0074] Step 5.1, the user inputs a question.

[0075] In the front-end form, enter "Which player performed best in the basketball games this July?", such as... Figure 2 As shown.

[0076] Step 5.2, construct the prompt parameter. The prompt parameter consists of role, task, data location, and specific requirements, as follows:

[0077] Character: You are a senior basketball research expert with over 20 years of research and experience.

[0078] Task: Analyze the performance of basketball players in the games this July and generate the name of the best performing player.

[0079] Data location: E:\Basketball Games\Game Statistics Data.xls

[0080] Specific requirements: 1) The best performing player will be displayed in text format, in a two-level format.

[0081] 2) Display player performance rankings in the form of charts.

[0082] Step 5.3: Utilize the reasoning capabilities of the large model to analyze the thinking process and generate a thought chain.

[0083] Step 6: Generate a workflow diagram.

[0084] Workflow diagram as follows Figure 3 The process includes six steps, in the following order: 1) Query the current date; 2) Query how many basketball games are scheduled for the corresponding date; 3) Query the players participating in each game; 4) Query the performance of each player on the court; 5) Calculate the overall performance of each player based on statistical indicators; 6) Determine the best player.

[0085] The Player Efficiency Rating (PER) from the NBA is used as a statistical indicator to quantify overall player performance. The formula is as follows:

[0086]

[0087] Positive contributions include points, rebounds, assists, steals, and blocks;

[0088] Negative contributions include turnovers, missed shots, and fouls.

[0089] The adjustment factor is used to standardize the data; it is calculated dynamically and depends on the team's real-time performance data.

[0090] Step 7: Generate node prompts.

[0091] Node prompts are generated from structured templates. A node prompt consists of three parts: description, function definition, and return line.

[0092] Step 8: Generate operation code.

[0093] Step 8.1: Based on the prompt for each operation, obtain the corresponding operation function.

[0094] Step 8.2: Send the requirements of each operation node to the LLM-Basketball model and call the helper.py function to generate the operation code.

[0095] Step 9: Perform the operation and return the final answer.

[0096] Step 9.1: Submit the workflow diagram to Operation and execute the operations in sequence.

[0097] The workflow diagram is divided into 6 steps, corresponding to 6 rounds of operations.

[0098] Step 9.2, Round 1 Operation: Submit the workflow diagram to Operation and execute Step 1 - Query the current date. Operation searches the database, obtains an execution result - the current date is July 31, 2025, and returns this result to the historical execution results, such as... Figure 3 As shown.

[0099] Operation Link submits the user's question, the first-step workflow, and the historical execution results to the Re-Step model to generate the second workflow.

[0100] The second workflow becomes: 1) Query how many basketball games are scheduled for July 2025; 2) Query the players participating in each game; 3) Query the on-court performance of each player; 4) Calculate the overall performance of each player based on statistical indicators; 5) Determine the best player.

[0101] Step 9.3, Round 2 Operation: Send the second workflow to Operation, and execute the first step of it—querying basketball games in July 2025. Obtain an execution result—there are 5 basketball games in July 2025—return this result, and add it to the historical execution results. Operation Link submits the user question, workflow, and historical execution results to the Re-Step model to generate the third workflow.

[0102] The third workflow becomes: 1) Query the players participating in each of the 5 matches; 2) Query the on-field performance of each participating player; 3) Calculate the overall performance of each player based on statistical indicators; 4) Determine the best player.

[0103] Round 3: The third workflow is sent to Operation, which executes the first step – querying the participating players for each of the 5 matches and returning the execution results, which are then added to the historical execution results. Operation Link will then submit the user question, workflow, and historical execution results to the Re-Step model to generate the fourth workflow.

[0104] The fourth workflow becomes: 1) Query the on-field performance of players such as Zhang San and Han Si; 2) Calculate the overall performance of each player based on statistical indicators; 3) Determine the best player.

[0105] Round 4: The fourth workflow is sent to Operation, which executes the first step – querying the on-field performance of players such as Zhang San and Han Si, and returning the execution results, which are then added to the historical execution results. Operation Link submits the user question, workflow, and historical execution results to the Re-Step model to generate the fifth workflow.

[0106] The fifth workflow becomes: 1) Calculate the overall performance of players such as Zhang San and Han Si according to the above PER formula; 2) Determine the best player.

[0107] Round 5: The fifth workflow is sent to Operation, which executes the first step – calculating the overall performance of players like Zhang San and Han Si based on the PER formula mentioned above, and returning the execution results, which are then added to the historical execution results. Operation Link submits the user question, workflow, and historical execution results to the Re-Step model to generate the sixth workflow.

[0108] The sixth workflow becomes: 1) Compare the overall performance values ​​to determine the best player.

[0109] Round 6: The sixth workflow is sent to Operation, which executes the first step – comparing overall performance metrics to determine the best player and returning the execution results, which are then added to the historical execution results. Operation Link submits the user's question, workflow, and historical execution results to the Re-Step model, returning the final answer.

[0110] Step 9.4: The task is complete, and the final answer is returned. The answer includes the name of the best player and a player performance ranking table.

[0111] In step 9.5, Operation Link displays the final answer to the user.

[0112] Step 10: The front end displays questions and answers and charts.

[0113] Step 10.1: Save the workflow as GraphML using the save_solution of the LLM-Basketball model and connect to the REST endpoint via FastAPI (interface).

[0114] Step 10.2: Transmit the names of the best-performing players and the player performance ranking table to the backend for API encapsulation. Both the player names and the player performance ranking table are in JSON format.

[0115] Step 10.3: The front-end calls the back-end interface and uses WebSocket for real-time interaction to transmit the data stream to the front-end.

[0116] Step 10.4: Render the player names and player performance ranking table based on the returned JSON data and display it on the front-end page.

[0117] This application also provides a basketball player evaluation and analysis system, including:

[0118] The first data acquisition module is used to collect athletes' exercise data, including physiological data, athletic ability data, physical indicator data, and recovery data.

[0119] The second data acquisition module is used to collect game statistics, including the basic rules and scoring methods of basketball games.

[0120] The preprocessing module is used to preprocess the motion data and the statistical data, and convert them to obtain data in the required format.

[0121] The storage module is used to store the converted data into a database based on the data source and type;

[0122] The parsing module is used to call the LLM-Basketball model based on the data in the database to generate workflow diagrams, node prompts, and operation codes;

[0123] The execution module is used to execute the operation code to obtain the final answer;

[0124] The presentation module is used for front-end Q&A and chart display.

[0125] The present invention also provides an electronic device, including at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the program is executed by the processing unit, the processing unit performs the above-described method.

[0126] The present invention also provides a storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the above-described method.

[0127] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating and analyzing basketball players, characterized in that, include: Collect athletes' exercise data, including physiological data, athletic ability data, physical indicators data, and recovery data; Collect statistical data on game techniques, including the basic rules and scoring methods of basketball; The motion data and statistical data are preprocessed and converted to obtain data in the required format. The converted data is stored in the database based on the data source and type. Based on the data in the database, the LLM-Basketball model is used to parse and generate workflow diagrams, node prompts, and operation codes. Executing the aforementioned operation code yields the final answer; The front end displays questions and answers as well as charts.

2. The basketball player evaluation and analysis method according to claim 1, characterized in that, The physiological data includes heart rate, TRIP, and calories; the exercise capacity data includes training load, training intensity, number of jumps, and number of changes of direction; the body indicator data includes age, height, weight, and BMI; and the recovery data includes recovery score and fatigue level.

3. The basketball player evaluation and analysis method according to claim 2, characterized in that, The statistical data includes 2 points scored, 3 points scored, free throws, fouls, turnovers, steals, and assists.

4. The basketball player evaluation and analysis method according to claim 3, characterized in that, Preprocessing of the motion data and the statistical data includes data cleaning, data calculation, and format conversion; Data cleaning refers to the process of processing and organizing data, including consistency checks, handling missing values, and removing duplicate data. The data calculation refers to converting the collected raw data into other data through calculation formulas; The format conversion refers to converting the data into the format required for subsequent analysis.

5. The basketball player evaluation and analysis method according to claim 4, characterized in that, Based on the data in the database, the LLM-Basketball model is used to parse and generate workflow diagrams, node prompts, and operation codes, including: Input a basketball-related question into the LLM-Basketball model; Based on the instantiated LLM-Basketball model, a prompt parameter is constructed, which consists of role, task, data location, and specific requirements for constructing the graph. The LLM-Basketball model's reasoning capabilities are invoked to analyze and generate thought processes; Execute the Python code to generate the mind chain, which is used to generate the workflow graph (work_graph); Execute the code to generate a work graph; Execute the work_graph code to generate a directed acyclic graph in GraphML format to represent the workflow; The node prompts are generated from a structured template and consist of three parts: description, function definition, and return line. The description is obtained from the node description in the work graph, the function definition is obtained based on the predecessor node, and the return line is obtained based on the successor node. Based on the prompt for each operation, the corresponding operation function is obtained. The operation function includes score statistics, player stability analysis, ranking statistics, shooting percentage prediction, and win rate prediction. Send the requirements of each operation node to the LLM-Basketball model, request code implementation, and generate node instructions; Based on the code template, the helper.py function of the LLM-Basketball model generates operation code.

6. The basketball player evaluation and analysis method according to claim 5, characterized in that, The final answer obtained by executing the above operation code includes: The execution operation is divided into 5 modules: User, Step Model, Re-Step Model, Operation, and OperationLink. The Step Model is responsible for generating the specific workflow, the Operation is responsible for executing each operation in the workflow, and the Operation Link is responsible for connecting the entire workflow. During operation, the workflow needs to be dynamically adjusted based on the execution results of each step. The Re-Step Model is responsible for modifying the workflow. The workflow diagram is submitted to the Operation, and the operations are executed sequentially according to the steps. The execution of the operations is divided into two stages: the execution stage and the Re-Step stage. After executing the first step of the workflow, the execution result of the first step is returned and placed into the historical execution results; the Operation Link submits the user issue, workflow, and historical execution results to the Re-Step model to generate a new workflow; after receiving the new workflow, the Operation Link executes the next operation; Repeat the above execution and Re-Step steps until all work steps are completed; Upon completion of the task, a final answer is returned, which may include questions and answers, tables, or diagrams. The Operation Link displays the final answer to the user.

7. The basketball player evaluation and analysis method according to claim 6, characterized in that, The front-end includes Q&A and chart display: The save_solution of the LLM-Basketball model is used to save the workflow as GraphML, and a REST endpoint is connected via FastAPI; The results are transmitted to the backend via API, and the API is encapsulated. The front-end calls the back-end interface and uses WebSocket for real-time interaction to achieve data stream transmission; After the front end obtains the data stream, it renders it using the echarts chart component library to achieve question-and-answer and chart display.

8. A basketball player evaluation and analysis system, characterized in that, include: The first data acquisition module is used to collect athletes' exercise data, including physiological data, athletic ability data, physical indicator data, and recovery data. The second data acquisition module is used to collect game statistics, including the basic rules and scoring methods of basketball games. The preprocessing module is used to preprocess the motion data and the statistical data, and convert them to obtain data in the required format. The storage module is used to store the converted data into a database based on the data source and type; The parsing module is used to call the LLM-Basketball model based on the data in the database to generate workflow diagrams, node prompts, and operation codes; The execution module is used to execute the operation code to obtain the final answer; The presentation module is used for front-end Q&A and chart display.

9. An electronic device, characterized in that, The method includes at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program that, when executed by the processing unit, causes the processing unit to perform the method described in any one of claims 1 to 7.

10. A storage medium, characterized in that, It stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the method described in any one of claims 1 to 7.

Images