Multi-data fusion-based user go chess power judgment system and method

Abstract

The invention relates to the technical field of artificial intelligence and go, in particular to a user go ability judgment system and method based on multi-data fusion. The core of the method is to construct an evaluation model comprising five dimensions of a historical chess playing winning rate, a decision thinking speed, an artificial intelligence chess piece falling coincidence rate, a layout strategy planning ability and an official receiving skill. And through multi-dimensional comprehensive evaluation, the Go level of the user is accurately quantified. An algorithm constructed based on the model supports three application scenes: firstly, an AI robot with equivalent chess power is matched for a user to play chess according to an evaluation result, and training challenging is ensured; secondly, identifying technical shortages of the user, and intelligently recommending a targeted exercise database; and 3, generating a standardized chess strength grade report according to the multi-dimensional score, and establishing a traceable technical growth file for the user.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and Go technology, and more specifically, to a system and method for judging a user's Go skill based on multi-data fusion. Background Art

[0002] Currently, Go player skill is primarily judged based on the traditional dan ranking system and the rating system used by online platforms. This traditional dan ranking system relies on offline tournaments, requiring players to spend months or even years participating in graded tournaments organized by Go associations. Furthermore, the difficulty level and judging standards of tournaments vary significantly across regions.

[0003] While online platforms use algorithms to dynamically adjust player ratings in real time, their core calculation model is still limited to the Elo algorithm's win-loss weighting system. However, Go is a complex intellectual sport, and judging skill solely by wins and losses fails to distinguish between luck and true skill. Ranks and ratings also fail to reflect differences in player ability at different stages of the game (e.g., opening, midgame, and endgame). Studies of professional Go players have shown a strong correlation between move time and decision-making quality. Moves with an average move time of less than 5 seconds have a 37% higher error rate than moves that are more carefully considered. Decision-making accuracy in complex situations is an even more crucial indicator of a player's overall strength.

[0004] Patent application number CN113935618B discloses a method, device, electronic device, and storage medium for evaluating chess playing ability. The method includes: first displaying a chessboard on an interface, searching and ranking recommended placement points based on a predetermined width; receiving user placement information and scoring it in conjunction with a win rate prediction model; then searching and determining the next move based on the board after the user's placement; and at the end of the game, generating an evaluation result based on the user's move score, enabling a rapid and accurate evaluation of chess skill. However, this method only considers the degree of consistency between the user's moves and the win rate prediction model, and does not consider the impact of the player's thinking time and decision-making efficiency at different stages on the accuracy of a player's chess skill assessment.

[0005] With the increase in the number of Go enthusiasts and the continuous improvement of Go competitive level, there is an urgent need to build a Go strength determination solution that can integrate multi-dimensional data to achieve a more scientific, comprehensive and accurate Go strength evaluation. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide a user Go chess strength judgment system and method based on multi-data fusion. By collecting and integrating multi-dimensional information such as the user's basic information, game history data, real-time game star position, etc., and using data analysis and machine learning algorithms, accurate and dynamic judgment of the user's Go chess strength can be achieved, providing a reliable chess strength evaluation basis for Go teaching, competition matching, etc.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A user Go skill judgment system based on multi-data fusion, including a multi-dimensional data collection and fusion module, a model training module, and a skill evaluation and dynamic update module;

[0009] Multi-dimensional data collection and fusion module: This module is responsible for collecting multi-dimensional game information of users, and deeply mining the features of each dimensional data, fusing them into a one-dimensional feature vector, so that each dimensional feature can better reflect the user's real chess ability. Based on historical game wins and losses and opponent level data, the user's scoring rate is calculated to reflect the user's performance when facing opponents of different levels; based on the position of the chess pieces and the thinking time, the chess game is divided into the layout, middle game and endgame stages, and the user's average thinking time is calculated in stages to reflect the user's ability performance in different stages; based on the AI-recommended moves, the matching rate between the user's decision and the AI's decision is calculated to reflect the breadth and depth of the chess player's thinking; based on the AI's real-time predicted winning rate trend, the value of the chess position is calculated to evaluate the user's ability to quickly determine the key move points and thus build a reasonable layout framework; based on the chess record data of the endgame stage, the endgame efficiency index is introduced to evaluate the user's efficiency and the rationality of the rhythm in the closing process;

[0010] Model Training Module: This module builds a random forest model to evaluate the user's Go skill. Based on the historical Go game data on the platform, it determines the tuning target and quantitatively calculates the user's overall Go skill by training and optimizing the Go skill evaluation model.

[0011] Chess strength evaluation and dynamic update module: This module calculates the user's comprehensive chess strength based on the user's historical chess record data, and automatically assigns weights based on the feature importance in the random forest model output. It generates a radar chart to show the strength and weakness distribution of the user's chess strength in five dimensions, compares the average level of the same level by dimension, and intelligently recommends targeted practice question banks for users. When generating new chess records, the module dynamically adjusts the user's single-dimensional chess strength and comprehensive chess strength based on data weighting.

[0012] A method for judging user Go skill based on multi-data fusion, the process is as follows:

[0013] Step 1: Multi-dimensional data collection and fusion.

[0014] Historical game win rate dimension: The Go game platform collects users' historical game data over a period of time, counts their win and loss results against opponents of different levels, and calculates their scoring rate.

[0015] Decision-making thinking speed dimension: The game platform obtains the user's thinking time for each move during the game. For situations of different difficulty levels (such as the opening, middle game, and endgame), the average thinking time is calculated and the global time distribution entropy is calculated.

[0016] Matching rate with AI: Calculate the average matching rate between user moves and AI moves. Effectively filter out inevitable moves that have no decision-making value, focus on key steps that reflect the player's proactive decision-making ability, and finely distinguish different levels of matching quality through a tiered scoring mechanism;

[0017] Layout strategic planning capability dimension: Calculate the average position value score S for the entire game location_value , helps measure the user's ability to quickly identify key placement points and thus build a reasonable layout framework;

[0018] Endgame closing technique: We obtained SGF-formatted game records from Go game platforms, analyzed the number of moves in the endgame, and used AI to evaluate the value of each move. We also introduced an endgame efficiency index to measure closing efficiency.

[0019] Multi-dimensional data fusion: All features of the five dimensions are integrated as input to the chess strength evaluation model, as follows:

[0020]

[0021] in, is the scoring rate, T avg_time is the global average thinking time, T opening_time is the average thinking time in the layout phase, T mid_time is the average thinking time in the middle game, T end_time is the average thinking time in the endgame, H(T) is the time distribution entropy, is the average score of the AI's move matching rate, S location_value is the global average positional value score, and EEI is the endgame efficiency index;

[0022] Step 2: Build and train chess strength evaluation model.

[0023] Data Acquisition: Game records of players of different skill levels are collected from the Go game platform. This ensures a balanced number of games played by players of different levels and types. The data mentioned in Step 1 is obtained and converted into features in five dimensions to form a dataset. Based on the Go platform's level classification, the corresponding Go skills are mapped from the lowest to the highest level into a range of 0 to 3000 points. This facilitates the numericalization of the skills and serves as a label for supervised learning.

[0024] Model Construction: Random Forest was chosen for Go skill assessment. Firstly, it has strong nonlinear fitting capabilities, suitable for handling Go's complex game characteristics and data patterns. Secondly, its feature importance is interpretable, making it easy to evaluate Go players' abilities in a dimensional manner within the model output.

[0025] Model training: The training and test sets are divided proportionally to ensure a balanced distribution of game data from Go players of different skill levels. The model is fitted using the training set, and parameters are tuned through cross-validation. The root mean square error (RMS) is used as the model evaluation metric.

[0026] Step 3: Chess strength assessment.

[0027] The player's most recent 100 game records are input into the Go skill evaluation model to calculate the user's average overall Go skill across those 100 games. A radar chart is generated based on the feature importance attributes in the output (quantifying the contribution of each dimension to overall Go skill), visually reflecting the user's Go skill across various dimensions. If a user's skill in a particular dimension is lower than the average for users of the same skill level on the Go platform, the system intelligently recommends targeted training content, such as learning AI layout patterns.

[0028] Step 4: Dynamic update of chess skills.

[0029] Every time a user plays a new game, the user's single-dimensional chess strength and comprehensive chess strength will be dynamically adjusted based on data weighting, and a technical growth file will be generated to record the changes in the user's chess strength.

[0030] Furthermore, the score rate in step 1 The specific calculation method is as follows:

[0031] The scores are weighted according to the opponent's chess strength level, using the formula Calculate the scoring rate, where n represents the total number of games counted, i = 1, 2, .., n, and the score of the i-th game is s i Using the formula Calculation, where α1 is the lower limit of the basic penalty. The smaller it is, the less points will be deducted when losing to a very strong opponent. α1∈[0,1]; α2 is the level difference coefficient. The larger it is, the more sensitive it is to the penalty. α3 is the losing streak penalty coefficient. The larger it is, the more severe the penalty for losing streak. α2, α3∈[0,1], α2 and α3 are associated with the user's current level. The higher the level, the more severe the penalty. u,i is the user level of the i-th game, r i is the level of the opponent in the i-th round, N lossis the current number of losing streaks; this scoring rate calculation method, through positive and negative scores and a dynamic penalty mechanism, can not only reward the performance of defeating strong opponents, but also effectively punish low-level mistakes. Compared with the traditional method of evaluating chess strength based solely on win rate data, it can more comprehensively reflect the user's true chess strength level and its fluctuations.

[0032] Furthermore, the calculation process of decision-making thinking speed in step 1 is as follows:

[0033] Using the formula Calculate the global draw thinking time to reflect the overall decision-making rhythm, where N represents the total number of steps in the game, T i Indicates the time of thinking at each step; using the formula Calculate the average thinking time in the layout phase and evaluate the decision efficiency in the layout phase, where P i Indicates the stage of the situation at step i (P i = {Planning, Middle Game, Endgame}), T i Indicates the thinking time of each step in the layout phase; using the formula Calculate the average thinking time in the middle game to reflect the thinking speed in complex situations, where T i Indicates the thinking time for each step in the middle game; using the formula Calculate the average thinking time in the endgame, which reflects the concentration and speed of the number of points calculation. i Indicates the thinking time for each step in the endgame; using the formula Calculate the time distribution entropy to measure the concentration of thinking time allocation. The lower the entropy value, the more time is concentrated on the key steps. N represents the total number of steps in the game, T i Indicates the time taken to think at each step.

[0034] Furthermore, the average score of the match rate with the AI ​​in step 1 is calculated as follows:

[0035] Obtain the valid move sequence from the Go platform analysis, expressed as S = {s1, s2, ..., s t}, where s t is the effective move for step t; for the chessboard at step t, we connect to open-source AI (such as Fine Art, KataGo, etc.) to generate a recommended list of K moves, which are expressed in order of priority as in For a one-dimensional index, use the formula Calculate the single-step score using the formula Calculate the average score of the rate of matching with the AI's moves.

[0036] Furthermore, the average position value score S in step 1 is location_valueThe calculation of , specifically includes the following process:

[0037] Use AI to analyze the layout stage of the chess game move by move, record the change in winning rate after each move, and use the formula Δwinrate i =winrate i -winrate i-1 , calculate the win rate impact value, where i is the number of hands and winrate0=50% is the initial win rate;

[0038] Use AI to calculate the potential for improving your win rate after placing a piece on each empty point, and select the maximum value Δwinrate i_max As a reference benchmark, combined with the actual calculated win rate impact value Δwinrate i , quantify each choice of the user in the layout stage, using the formula Calculate the average positional value score for the entire game.

[0039] Furthermore, the calculation process of the endgame efficiency index in step 1 is as follows:

[0040] The endgame efficiency index represents the ratio of the average number of chess moves made by a chess player in each step of the endgame to the theoretical optimal value, and is then corrected by combining the number of mistakes and the rationality of the rhythm. Calculate the endgame efficiency index, where n represents the number of effective moves in the endgame phase; V user,i V represents the actual number of official moves made by the user in step i; ai,i is the optimal number of official moves recommended by AI for step i; W i It is the order weight, which is adjusted dynamically to reflect the influence of the order of moves on the number of points of the official piece. It usually decreases with the progress. When the "reversed size" move occurs (V user,i >V user,i+1 ), so that W i <1; E is the number of mistakes in the closing stage (e.g., the value of the move is less than 50% of the AI ​​recommended value); R∈[0,1] is the rhythm correction coefficient, which measures the rationality of the decay of the move value. The larger the value, the more reasonable the closing rhythm. Using the formula calculate.

[0041] Furthermore, in step 4, the chess strength is dynamically updated. The specific process is as follows:

[0042] For each dimension i (i = 1, 2, 3, 4, 5), a weighted average combined with time decay method is used, using the formula Calculate and update chess strength, where is the chess strength of the i-th dimension after the t-th update, represents the chess strength of the i-th dimension calculated by the chess strength evaluation model in the new game. α is the weight of historical data, reflecting the credibility of historical scores. β is the weight of new data, reflecting the credibility of new game scores. It can be dynamically adjusted according to the opponent's chess strength to avoid excessive fluctuations in chess strength due to extreme games. To prevent historical data from excessively affecting recent scores, a time decay factor γ (0<γ<1) is introduced: α (t+1) =γ×α (t) , as time goes by, the weight of historical data decays exponentially, making the model pay more attention to recent performance; accordingly, using the formula Calculate the user's current comprehensive chess strength.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. High Accuracy: Breaking through the traditional single-win or rating evaluation model, this system integrates multi-dimensional data, including historical game win rate, decision-making speed, match rate with AI moves, strategic planning ability, and endgame and closing techniques. Each dimension is quantified through a refined algorithm, covering all stages of Go ability. Compared to traditional single-dimensional Go skill assessment methods, this system comprehensively captures the differences in user performance at different stages and can more accurately quantify and reflect the user's true Go skill level.

[0045] 2. Refined dynamic evaluation: Relying on the importance of features output by the random forest model, the system automatically assigns weights to the five dimensions, generates a visual radar chart, and compares the average levels of the same level by dimension, achieving a breakthrough from comprehensive scoring to single-dimensional ability analysis. Users can clearly identify their specific shortcomings in each stage of Go and use the exercises recommended by the system to carry out targeted improvement. At the same time, a weighted average combined with a time decay algorithm is introduced. After each game, the user automatically updates the single-dimensional and comprehensive Go strength, generates a traceable technical growth profile, and helps form a closed-loop optimization path of "assessment-positioning-improvement-reassessment". BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a principle block diagram of a user Go skill judgment system based on multi-data fusion according to the present invention;

[0047] Figure 2 This is a method flow chart of a user Go skill judgment system based on multi-data fusion according to the present invention. DETAILED DESCRIPTION

[0048] Example 1, refer to Figure 1 In this embodiment, a user Go chess strength judgment system based on multi-data fusion includes a multi-dimensional data collection and fusion module, a model training module, and a chess strength evaluation and dynamic update module.

[0049] The multi-dimensional data collection and fusion module is responsible for collecting user game information. It analyzes the game records recorded by the Go platform to obtain historical game wins and losses, opponent level data, move positions and thinking time, the impact of moves on AI-recommended moves, and the AI's real-time predicted win rate trends. It then deeply mines the characteristics of each dimension of data and fuses them into a one-dimensional feature vector, making each dimension feature more reflective of the user's actual Go ability. Based on historical game wins and opponent level data, it calculates the user's scoring rate to reflect the user's performance against opponents of different levels. Based on move positions and thinking time, the game is divided into the layout, middle game, and endgame stages, and the user's average thinking time is calculated for each stage to reflect the user's performance at each stage. Based on the impact of moves on AI-recommended moves, the consistency rate between user decisions and AI decisions is calculated to reflect the breadth and depth of the player's thinking. Based on the AI's real-time predicted win rate trends, the value of move positions is calculated to evaluate the user's ability to quickly determine key moves and thus build a reasonable layout framework. Based on the game record data of the endgame stage, an endgame efficiency index is introduced to evaluate the user's efficiency and the rationality of the rhythm during the endgame.

[0050] The model training module builds a random forest model to evaluate the user's Go skill. Based on the historical Go game data on the platform, it determines the tuning target and quantitatively calculates the user's overall Go skill by training and optimizing the Go skill evaluation model.

[0051] The chess strength evaluation and dynamic update module calculates the user's comprehensive chess strength based on the user's historical chess record data, and automatically assigns weights based on the feature importance in the random forest model output. It generates a radar chart to show the strength and weakness distribution of the user's chess strength in five dimensions, compares the average level of the same level by dimension, and intelligently recommends targeted practice question banks for users. When generating new chess game records, the module dynamically adjusts the user's single-dimensional chess strength and comprehensive chess strength based on data weighting.

[0052] Example 2, refer to Figure 2 The method for judging a user's Go skill based on multi-data fusion in this embodiment specifically includes the following steps:

[0053] Step 1: Multi-dimensional data collection and fusion.

[0054] Historical game win rate dimension: The Go game platform collects users' historical game data over a period of time, counts their win and loss results against opponents of different levels, and calculates their scoring rate. The scores are weighted according to the opponent's chess strength level. The specific calculation formula is as follows:

[0055]

[0056] Where n represents the total number of games counted, i = 1, 2, .., n, si is the score of the i-th game, which is calculated as follows:

[0057]

[0058] Among them, α1 is the lower limit of the basic penalty. The smaller the value, the less points will be deducted if losing to a very strong opponent. α1∈[0,1], in this embodiment, α1=0.8; α2 is the level difference coefficient. The larger the value, the more sensitive it is to the penalty. α3 is the losing streak penalty coefficient. The larger the value, the more severe the penalty for losing streak. α2, α3∈[0,1], α2 and α3 are associated with the user's current level. The higher the level, the more severe the penalty. In this embodiment, α2=0.5, α3=0.5; r u,i is the user level of the i-th game, r i is the level of the opponent in the i-th round, N loss is the current number of losing streaks; this scoring rate calculation method, through positive and negative scores and a dynamic penalty mechanism, can not only reward the performance of defeating strong opponents, but also effectively punish low-level mistakes. Compared with the traditional method of evaluating chess strength based solely on win rate data, it can more comprehensively reflect the user's true chess strength level and its fluctuations.

[0059] Decision-making thinking speed dimension: The game platform is used to obtain the user's thinking time for each move during the game. For situations of different difficulty levels (such as the opening, middle game, and endgame), the average thinking time and the global time distribution entropy are calculated.

[0060] Global average thinking time T avg_time , reflecting the overall decision-making rhythm, the calculation formula is as follows:

[0061]

[0062] Among them, N represents the total number of steps in the game, T i Indicates the time of thinking at each step;

[0063] Average thinking time T in the layout phase opening_time , to evaluate the decision efficiency in the layout phase, the specific formula is as follows:

[0064]

[0065] Among them, P i Indicates the stage of the situation at step i (P i = {Planning, Middle Game, Endgame}), T i Indicates the thinking time for each step in the layout phase;

[0066] Average thinking time T in the middle game mid_time , reflecting the speed of thinking in complex situations. The specific formula is as follows:

[0067]

[0068] Among them, T i Indicates the thinking time for each move in the middle game;

[0069] Average thinking time in the endgame T end_time , reflecting the concentration and speed of mesh number calculation. The specific formula is as follows:

[0070]

[0071] Among them, T i Indicates the thinking time for each move in the endgame;

[0072] The time distribution entropy H(T) measures the concentration of thinking time allocation. The lower the entropy value, the more time is concentrated on the key steps. The calculation formula is as follows:

[0073]

[0074] Among them, N represents the total number of steps in the game, T i Indicates the time of thinking at each step;

[0075] The consistency of the move with the AI ​​dimension: Get the SGF format game record from the Go platform, parse the coordinates (x, y) of each move, and convert it into a one-dimensional index t: t = y × N + x, where N is the length of the board; extract the valid move sequence from the SGF file, excluding the necessary moves (such as the move of taking the piece, the move of robbing, etc.), and the valid move sequence is expressed as S = {s1, s2, ..., s t}, where s t is the effective move for step t; for the chessboard at step t, we connect to open-source AI (such as Fine Art, KataGo, etc.) to generate a recommended list of K moves, which are expressed in order of priority as in is a one-dimensional index, and the corresponding single-step score can be expressed as follows:

[0076]

[0077] Furthermore, the formula for calculating the average score of the AI's move matching rate is as follows:

[0078]

[0079] Through this calculation method, we can effectively filter out inevitable moves that have no decision-making value, focus on key steps that reflect the player's active decision-making ability, and use a step-by-step scoring mechanism to finely distinguish different levels of matching quality, providing more accurate quantitative indicators for chess strength evaluation.

[0080] Layout strategic planning capability dimension: win rate impact value Δwinrate i , use AI to analyze the layout stage of the chess record (such as the first 30 or 50 moves) move by move, and record the changes in the winning rate after each move. The calculation formula is as follows:

[0081] Δwinrate i =winrate i -winrate i-1 ;

[0082] Where i is the number of lots, winrate0=50% is the initial winning rate;

[0083] Positional value is a quantitative indicator that evaluates the importance of each empty point on the chessboard to the current chess game. AI can conduct a comprehensive and in-depth analysis of the chessboard situation, including the position of the pieces, their connections, luck, potential development space, etc. AI calculates the potential for improving the winning rate of the team after placing a piece on each empty point, and selects the maximum value Δwinrate i_max As a reference benchmark, combined with the actual calculated win rate impact value Δwinrate i , quantify each choice of the user in the layout phase (out of 10 points), and calculate the average position value score of the entire game. The specific calculation method is as follows:

[0084]

[0085] During the Go layout stage, the calculation of position value can help measure the user's ability to quickly determine key placement points and thus build a reasonable layout framework.

[0086] Endgame closing technique: Obtaining SGF-formatted game records from Go game platforms, analyzing the number of moves in the endgame, and using AI to assess the value of each move. The endgame efficiency index is then calculated to measure closing efficiency.

[0087] The Endgame Efficiency Index (EEI) is introduced to represent the ratio of the average number of chess moves made by a player in each endgame move to the theoretical optimal value, and is then corrected based on the number of mistakes and the rationality of the rhythm. The calculation formula is as follows:

[0088]

[0089] Where n represents the number of valid moves in the endgame; V user,i V represents the actual number of official moves made by the user in step i; ai,i is the optimal number of official moves recommended by AI for step i; W iIt is the order weight, which is adjusted dynamically to reflect the influence of the order of moves on the number of points of the official piece. It usually decreases with the progress. When the "reversed size" move occurs (V user,i >V user,i+1 ), so that W i <1; E is the number of mistakes in the closing phase (e.g., the value of a move is less than 50% of the AI-recommended value); R∈[0,1] is the rhythm correction coefficient, which measures the rationality of the decay of the value of a move. A larger value indicates a more reasonable closing rhythm. The specific calculation method is as follows:

[0090]

[0091] Multi-dimensional data fusion: All features of the five dimensions are integrated as input to the chess strength evaluation model, as follows:

[0092]

[0093] in, is the scoring rate, T avg_time is the global average thinking time, T opening_time is the average thinking time in the layout phase, T mid_time is the average thinking time in the middle game, T end_time is the average thinking time in the endgame, H(T) is the time distribution entropy, is the average score of the AI's move matching rate, S location_value is the global average positional value score, and EEI is the endgame efficiency index;

[0094] Step 2: Build and train chess strength evaluation model.

[0095] Data Acquisition: Game records of players of different skill levels are collected from the Go game platform. This ensures a balanced number of games played by players of different levels and types. The data mentioned in Step 1 is obtained and converted into features in five dimensions to form a dataset. Based on the Go platform's level classification, the corresponding Go skills are mapped from the lowest to the highest level into a range of 0 to 3000 points. This facilitates the numericalization of the skills and serves as a label for supervised learning.

[0096] Model Construction: Random Forest has unique advantages in Go skill assessment. First, it has strong nonlinear fitting capabilities, suitable for handling Go's complex game characteristics and data patterns. Second, feature importance is interpretable, making it easy to evaluate Go players' abilities in a dimensional manner in the model output. In this embodiment, the parameters of the random forest model are selected as follows:

[0097] The number of decision trees is 500. A sufficient number ensures that nonlinear relationships can be captured while improving model robustness and preventing overfitting of a single tree. The maximum depth of a single tree is 15, balancing model complexity and interpretability. Too deep a tree makes it difficult to interpret feature importance by dimension. The minimum number of samples for node splitting is 5, which controls the node splitting threshold and avoids interference from abnormal data. It is suitable for scenarios with many feature dimensions. The minimum number of samples for leaf nodes is 3, which prevents the model from learning local noise. When the sample size is large, it can be adjusted to a smaller value. The number of features selected for each split is 'sqrt', which is the default value. The number of random seeds is fixed to ensure reproducibility of the experiment.

[0098] Model training: The training set and test set are divided into a ratio of 7:3 to ensure a balanced distribution of game data of Go players with different levels. The training set is used to fit the model, and the parameters are tuned through cross-validation (five-fold cross-validation). The model evaluation indicators are the root mean square error and the coefficient of determination R. 2 , requiring the test set R 2 ≥0.85.

[0099] Step 3: Chess strength assessment.

[0100] The chess game data of the player's most recent 100 games are input into the chess strength evaluation model to automatically calculate the user's average comprehensive chess strength in the past 100 games. When matching opponents, users with similar comprehensive chess strength are matched. At the same time, a radar chart is generated based on the feature importance attributes in the output (quantifying the contribution of each dimension to the total chess strength) to intuitively reflect the user's chess strength in each dimension. If a dimension of the user is lower than the average chess strength of users with the same chess strength on the Go platform, the system intelligently recommends targeted training content, such as learning AI layout patterns.

[0101] Step 4: Dynamic update of chess skills.

[0102] Each time a user plays a new game, their single-dimensional chess skill and overall chess skill will be dynamically adjusted based on data weighting, and a skill growth profile will be generated to record the user's chess skill changes. For each dimension i (i = 1, 2, 3, 4, 5), the chess skill is updated using a weighted average combined with time decay. The specific calculation formula is as follows:

[0103]

[0104] in, is the chess strength of the i-th dimension after the t-th update, represents the chess strength of the i-th dimension calculated by the chess strength evaluation model in the new game. α is the weight of historical data, reflecting the credibility of historical scores. In this embodiment, α = 0.8. β is the weight of new data, reflecting the credibility of new game scores. It can be dynamically adjusted according to the opponent's chess strength to avoid excessive fluctuations in chess strength due to extreme games. In this embodiment, β = 0.9. To prevent historical data from excessively affecting recent scores, a time decay factor γ (0 < γ < 1) is introduced: α (t+1) =γ×α (t) , as time goes by, the weight of historical data decays exponentially, making the model pay more attention to recent performance;

[0105] The comprehensive chess strength is obtained by weighted summation of each dimension, and the calculation method is as follows:

[0106]

[0107] in, Represents the weight of the i-th dimension after the t+1th update;

[0108] Through the detailed introduction of the above embodiments, the present invention's system and method for determining a user's Go skill based on multi-data fusion deeply explores deep features across five dimensions, enabling it to better reflect the user's true Go skill. An evaluation model has been constructed that encompasses five dimensions: historical game win rate, decision-making and thinking speed, match rate with artificial intelligence (AI), strategic planning capabilities, and endgame closing techniques. Through a comprehensive multi-dimensional assessment, the weight of each dimension is automatically analyzed to accurately quantify the user's Go skill. Based on this comprehensive assessment of the user's Go skill, this evaluation model calculates the specific skill level for each dimension and, based on this, intelligently recommends a targeted practice question bank to help the user address shortcomings.

[0109] The above formulas are all dimensionless and numerically calculated, and the preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0110] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0111] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0112] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0113] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0114] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0115] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0116] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for judging user Go skill based on multi-data fusion, characterized in that: The method flow is as follows: Step 1: Multi-dimensional data collection and integration: Collect historical data on five dimensions: the user's historical game win rate, decision-making speed, match rate with the AI, strategic planning ability, and endgame closing skills. Perform preliminary feature mining and then integrate them into a one-dimensional feature vector. Step 2: Build and train a chess skill evaluation model: Select appropriate parameters to build a random forest model and train the model using the one-dimensional feature vector fused in step 1. Step 3: Chess skill evaluation: Use the chess skill evaluation model from step 2 to evaluate the user's chess skill; Step 4: Dynamic update of chess strength: For newly generated game data, use the chess strength evaluation model to dynamically adjust the user's single-dimensional chess strength and comprehensive chess strength.

2. The method for judging a user's Go skill based on multi-data fusion according to claim 1, characterized in that: The score rate described in step 1 The specific method for calculating is: First, the scores are weighted according to the opponent's chess strength level. The score s of the i-th game is i Using the formula Calculate, where α1 is the lower limit of the basic penalty, α2 is the level difference coefficient, α3 is the consecutive defeat penalty coefficient, α1, α2, α3∈[0,1]; r u,i is the user level of the i-th game, r i is the level of the opponent in the i-th round, N loss is the current number of losing streaks, and then the scoring rate is calculated based on the score of a single game.

3. The method for judging a user's Go skill based on multi-data fusion according to claim 1, characterized in that: The specific calculation process of the decision-making thinking speed described in step 1 includes: Calculate the global average thinking time, the average thinking time in the layout phase, the average thinking time in the middle game phase, the average thinking time in the endgame phase, and the time distribution entropy respectively.

4. The method for judging a user's Go skill based on multi-data fusion according to claim 1, characterized in that: The average score of the AI's match rate in step 1 is calculated as follows: Obtain the valid move sequence from the Go platform analysis, expressed as S = {s1, s2, ..., s t }, where s t is the effective move for step t; for the chessboard at step t, we connect to the open source AI to generate a recommended list of K moves, which are expressed in order of priority as in It is a one-dimensional index. The single-step score is calculated using a step-by-step scoring method, and the average score is calculated.

5. The method for judging a user's Go skill based on multi-data fusion according to claim 1, characterized in that: The average positional value score for the entire game, as described in step 1, is calculated as follows: Use AI to analyze the layout stage of the chess game move by move, record the change in winning rate after each move, and use the formula Δwinrate i =winrate i -winrate i-1 , calculate the win rate impact value, where i is the number of hands and winrate0=50% is the initial win rate; AI is used to calculate the potential for improving one's own winning rate after placing a piece on each empty point, and the maximum value is selected as a reference benchmark; combined with the actual calculated winning rate impact value, the user's choice of each hand in the layout phase is quantified, and the average position value score for the entire game is calculated.

6. The method for judging a user's Go skill based on multi-data fusion according to claim 1, characterized in that: The endgame efficiency index described in step 1 is calculated as follows: Using the formula Calculate the endgame efficiency index, where n represents the number of effective moves in the endgame phase; V user,i V represents the actual number of official moves made by the user in step i; ai,i is the optimal number of official moves recommended by AI for step i; W i is the order weight, when V user,i >V user,i+1 , so that W i <1; E is the number of mistakes in the final stage; R∈[0,1] is the rhythm correction coefficient, using the formula calculate.

7. The method for judging a user's Go skill based on multi-data fusion according to claim 1, characterized in that: The dynamic update of chess strength described in step 4 includes the following steps: For each dimension i, i = 1, 2, 3, 4, 5, a weighted average combined with time decay method is used, using the formula Calculate and update chess strength, where is the chess strength of the i-th dimension after the t-th update, represents the chess strength of the i-th dimension calculated by the chess strength evaluation model in the new game, α is the weight of historical data, β is the weight of new data; the time decay factor γ is introduced: α (t+1) =γ×α (t) , as time goes by, the weight of historical data decays exponentially; accordingly, the weight corresponding to each dimension is used to calculate the user's current comprehensive chess ability.

8. A system for judging user Go skill based on multi-data fusion, for implementing a method for judging user Go skill based on multi-data fusion as claimed in any one of claims 1 to 7, characterized in that: The system includes a multi-dimensional data collection and fusion module, a model training module, and a chess strength evaluation and dynamic update module; Multi-dimensional data collection and fusion module: This module is responsible for collecting multi-dimensional game information from users, deeply mining the characteristics of each dimension of data, and fusing them into a one-dimensional feature vector. It calculates the user's scoring rate based on historical game wins and losses and opponent level data. It divides the game into layout, midgame, and endgame stages based on move positions and thinking time, and calculates the user's average thinking time in each stage. It calculates the consistency rate between user decisions and AI decisions based on the AI's recommended moves. It calculates the value of move positions based on the AI's real-time prediction of win rate trends. It introduces and calculates the endgame efficiency index based on the chess record data of the endgame stage. Model Training Module: This module builds a random forest model, determines optimization targets based on historical Go platform game data, and trains and optimizes the Go skill evaluation model using datasets acquired through the Multi-Dimensional Data Collection and Fusion Module, achieving quantitative calculation of the user's comprehensive Go skill. Chess strength evaluation and dynamic update module: This module calculates the user's comprehensive chess strength based on the user's historical chess data using the chess strength evaluation model trained by the model training module. It automatically assigns weights based on the importance of the features in the model output, and generates a radar chart to display the strength and weakness distribution of the user's chess strength in five dimensions. It compares the average level of the same level by dimension and intelligently recommends targeted practice question banks to users. When generating new chess game records, the user's single-dimensional chess strength and comprehensive chess strength are dynamically adjusted based on data weighting.

Images