Generation system and method of explanation cue word

Abstract

The invention provides a system and method for generating explanation cues, and the system comprises an environment state observation and feature extraction module which is used for carrying out the collection and feature extraction of the information of an original observable environment, and obtaining a basic data set; the intelligent agent observable environment state feature extraction module is used for generating a playing combination according to the basic data set and analyzing the current situation to obtain an environment state feature set; and the text generation prompt module is used for integrating the multi-source information and generating explanation prompt words. According to the technical scheme, through an automatic generation mechanism, high-quality explanation cues are quickly generated by utilizing an efficient algorithm and powerful computing resources; manual intervention is reduced, and the production efficiency is greatly improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a system and method for generating commentary prompt words. Background Art

[0002] Under the current technical environment, the methods for generating commentary prompt words are mainly divided into two types: template method and manual generation.

[0003] Template-based approach: This method generates prompts by filling relevant information into a pre-defined template. Its advantages lie in its straightforward implementation and low development costs. However, when faced with complex and ever-changing real-world needs, the template-based approach has significant limitations. Because the templates are fixed, the generated prompts lack personalized features and cannot be flexibly adjusted to suit individual user preferences and specific contexts, resulting in a lack of specificity and adaptability.

[0004] Manual generation: Professional human experts manually create prompts based on their knowledge and experience. This method ensures the quality of prompts and often offers superior accuracy, professionalism, and linguistic expression. However, manual generation is extremely inefficient. As business scales, it requires significant manpower and time investment, making scalability difficult and costs escalate dramatically.

[0005] The problems and shortcomings of the existing technology are mainly reflected in the following aspects: Lack of personalization: Most existing generation methods generate generic prompts that lack close connection to users' personal preferences, usage scenarios, and real-time context. They fail to meet the diverse and personalized needs of different users in different scenarios, resulting in a poor user experience.

[0006] Lack of flexibility: Template-based approaches exhibit poor adaptability when dealing with complex scenarios or dynamically changing situations. When requirements change or new scenarios emerge, it is difficult to quickly adjust and optimize the templates, making it difficult to quickly respond to new requirements and changes.

[0007] Low efficiency of manual generation: Manually writing prompts is time-consuming and labor-intensive. This method is clearly inadequate in scenarios where a large number of commentary prompts need to be produced quickly, such as real-time live commentary and large-scale content generation, severely restricting business development and expansion. Summary of the Invention

[0008] The present application provides a system and method for generating commentary prompt words, so as to improve the quality and efficiency of commentary prompt word generation.

[0009] In a first aspect, a system for generating commentary prompt words is provided, comprising: The environmental state observation and feature extraction module is used to collect and extract features from the original observable environment information to obtain the basic data set; The agent can observe the environment state feature extraction module, which is used to generate a card combination based on the basic data set and analyze the current situation to obtain the environment state feature set; Generate text prompt module, which is used to integrate multi-source information and generate explanation prompt words.

[0010] In the above technical solution, an environmental state observation and feature extraction module is set up to collect and extract features from the original observable environment to obtain a basic data set; an intelligent agent observable environmental state feature extraction module is used to generate a card-playing combination based on the basic data set and analyze the current situation to obtain an environmental state feature set; a text prompt generation module is used to integrate multi-source information and generate commentary prompts; through an automated generation mechanism, efficient algorithms and powerful computing resources are utilized to quickly generate high-quality commentary prompts; reducing manual intervention and greatly improving production efficiency.

[0011] In a specific embodiment, it also includes: The memory storage and update mechanism module is used to establish an index structure for the game timeline and completely save the card-playing records of the entire game; and by establishing a player status model, it monitors its own status characteristics in real time; and obtains the player's operation data and status information in real time.

[0012] In a specific embodiment, it also includes: The card-playing action possibility legality calculation and coding serialization module is used to calculate all possible card plays in the current round; generate all card-playing combinations that comply with the rules; and code and serialize each possible card play.

[0013] In a specific implementation scheme, the environmental state observation and feature extraction module includes: Multi-source information combination submodule, used to integrate multi-source information including RGB images and text information; The streaming data feature parsing and extraction submodule is used to assist the intelligent agent in aligning the information of the original observable environment with the real-time status observation data of the intelligent agent represented in text format through the adaptive template generation technology based on deep learning; the RGB image feature parsing and extraction submodule is used to use computer vision technology to locate and capture information when using the RGB image of the card-playing screen as the carrier to obtain the hand and card-playing information.

[0014] In a specific implementation scheme, the agent-observable environment state feature extraction module includes: The card playing combination generation unit is used to generate all optional card playing combinations in this card playing by using rule reasoning and algorithm generation technology.

[0015] In a specific implementation scheme, the memory storage and update mechanism module includes: The game progress recording unit is used to establish an index structure for the game timeline and fully save the card play records of the entire game; The player status tracking unit is used to monitor the player's own status characteristics in real time by establishing a player status model; and to obtain the player's operation data and status information in real time.

[0016] In a specific embodiment, the card playing probability calculation and coding serialization module includes: The card playing possibility calculation unit is used to calculate all possible card playing possibilities in the current round and generate all card playing combinations that meet the rules; The coding serialization unit is used to code and serialize each possible card played.

[0017] In a specific embodiment, the generating text prompt module includes: Multi-source information integration unit, used to deeply fuse RGB images and streaming data to obtain comprehensive environmental information; The text prompt generation unit is used to generate commentary prompt words based on the comprehensive environmental information using natural language generation technology and deep learning models.

[0018] In a second aspect, a method for generating commentary prompt words is provided, comprising the following steps: The environment state observation and feature extraction module is used to collect and extract features from the original observable environment to obtain a basic data set; Utilizing the agent-observable environment state feature extraction module to generate a card-playing combination based on the basic data set, and analyzing the current situation to obtain an environment state feature set; The text prompt generation module is used to integrate multi-source information and generate commentary prompts.

[0019] In a specific embodiment, it also includes: The memory storage and update mechanism module is used to establish an index structure for the game timeline, completely preserving the entire game's card-playing records; and by establishing a player state model, the player's own state characteristics are monitored in real time; the player's operation data and state information are obtained in real time; the card-playing action possibility legality calculation and coding serialization module is used to calculate all possible card plays in the current round; all card-playing combinations that comply with the rules are generated; and each possible card-playing combination is coded and serialized.

[0020] In the above technical solution, an environmental state observation and feature extraction module is set up to collect and extract features from the original observable environment to obtain a basic data set; an intelligent agent observable environmental state feature extraction module is used to generate a card-playing combination based on the basic data set and analyze the current situation to obtain an environmental state feature set; a text prompt generation module is used to integrate multi-source information and generate commentary prompts; through an automated generation mechanism, efficient algorithms and powerful computing resources are utilized to quickly generate high-quality commentary prompts; reducing manual intervention and greatly improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A structural block diagram of a system for generating explanation prompt words provided in an embodiment of the present application; Figure 2 This is a flowchart of a method for generating explanation prompt words provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] The present application will be further described in detail below with reference to the accompanying drawings and examples, through which the features and advantages of the present application will become more clearly understood.

[0023] The word "exemplary" is used exclusively herein to mean "serving as an example, example, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0024] In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0025] To facilitate understanding of the system and method for generating commentary prompts provided in the embodiments of this application, let's first explain their application scenarios. The system and method for generating commentary prompts provided in the embodiments of this application are designed to improve the quality and efficiency of commentary prompt generation. Problems and shortcomings of the prior art are primarily reflected in the following aspects: 1. Lack of personalization: Most existing generation methods generate commentary prompts that are uniform and lack close connection to the user's personal preferences, usage scenarios, and real-time context. This fails to meet the diverse and personalized needs of different users in different scenarios, resulting in a poor user experience. 2. Lack of flexibility: Template-based methods exhibit extremely poor adaptability when dealing with complex scenarios or dynamically changing situations. When requirements change or new scenarios emerge, it is difficult to quickly adjust and optimize the templates, making it difficult to quickly respond to new requirements and changes. 3. Low manual generation efficiency: Manually writing commentary prompts is time-consuming and labor-intensive. In scenarios where a large number of commentary prompts need to be produced quickly, such as live commentary and large-scale content generation, this method clearly fails to meet the requirements, severely limiting business development and expansion. To this end, the present invention provides a system and method for generating explanation prompt words to improve the quality and efficiency of generating explanation prompt words.

[0026] refer to Figure 1 and Figure 2 , Figure 1 A structural block diagram of a system for generating explanation prompt words provided in an embodiment of the present application; Figure 2 This is a flowchart of a method for generating explanation prompt words provided in an embodiment of the present application.

[0027] exist Figure 1 In the embodiment of the present application, a system for generating explanation prompt words is provided, including: The environmental state observation and feature extraction module is used to collect and extract features from the original observable environment information to obtain the basic data set; The agent can observe the environment state feature extraction module, which is used to generate a card combination based on the basic data set and analyze the current situation to obtain the environment state feature set; Generate text prompt module, which is used to integrate multi-source information and generate explanation prompt words.

[0028] In the above technical solution, an environmental state observation and feature extraction module is set up to collect and extract features from the original observable environment to obtain a basic data set; an intelligent agent observable environmental state feature extraction module is used to generate a card-playing combination based on the basic data set and analyze the current situation to obtain an environmental state feature set; a text prompt generation module is used to integrate multi-source information and generate commentary prompts; through an automated generation mechanism, efficient algorithms and powerful computing resources are utilized to quickly generate high-quality commentary prompts; reducing manual intervention and greatly improving production efficiency.

[0029] In a specific embodiment, it also includes: The memory storage and update mechanism module is used to establish an index structure for the game timeline and completely save the card-playing records of the entire game; and by establishing a player status model, it monitors its own status characteristics in real time; and obtains the player's operation data and status information in real time.

[0030] In a specific embodiment, it also includes: The card-playing action possibility legality calculation and coding serialization module is used to calculate all possible card plays in the current round; generate all card-playing combinations that comply with the rules; and code and serialize each possible card play.

[0031] In a specific implementation scheme, the environmental state observation and feature extraction module includes: Multi-source information combination submodule, used to integrate multi-source information including RGB images and text information; The streaming data feature parsing and extraction submodule is used to assist the intelligent agent in aligning the information of the original observable environment with the real-time status observation data of the intelligent agent represented in text format through the adaptive template generation technology based on deep learning; the RGB image feature parsing and extraction submodule is used to use computer vision technology to locate and capture information when using the RGB image of the card-playing screen as the carrier to obtain the hand and card-playing information.

[0032] In a specific implementation scheme, the agent-observable environment state feature extraction module includes: The card playing combination generation unit is used to generate all optional card playing combinations in this card playing by using rule reasoning and algorithm generation technology.

[0033] In a specific implementation scheme, the memory storage and update mechanism module includes: The game progress recording unit is used to establish an index structure for the game timeline and fully save the card play records of the entire game; The player status tracking unit is used to monitor the player's own status characteristics in real time by establishing a player status model; and to obtain the player's operation data and status information in real time.

[0034] In a specific embodiment, the card playing probability calculation and coding serialization module includes: The card playing possibility calculation unit is used to calculate all possible card playing possibilities in the current round and generate all card playing combinations that meet the rules; The coding serialization unit is used to code and serialize each possible card played.

[0035] In a specific embodiment, the generating text prompt module includes: Multi-source information integration unit, used to deeply fuse RGB images and streaming data to obtain comprehensive environmental information; The text prompt generation unit is used to generate commentary prompt words based on the comprehensive environmental information using natural language generation technology and deep learning models.

[0036] Specifically, the system for generating the commentary prompt words includes: 1. Environmental state observation and feature extraction module, including: Multi-source information combination submodule: Advanced information fusion technology is used to organically integrate information from multiple sources, including RGB images and text. Image recognition and analysis algorithms are used to extract key features from images, such as the shape, color, and numbers of the cards. For text, natural language processing techniques are employed to perform lexical analysis, syntactic analysis, and semantic understanding, extracting key information and semantic features from the text. Subsequently, a unified data structure and feature representation method are established to fuse these different types of information, providing a comprehensive data foundation for subsequent analysis.

[0037] Streaming data feature analysis and extraction sub-module: For the real-time state observation data of the intelligent agent represented in text formats such as JSON and XML, the adaptive template generation technology based on deep learning is used to assist the intelligent agent in aligning the information of the original observable environment. The specific steps are as follows: The seat number accurate extraction unit is used to extract the seat number ID (0-3) of the current card played and clarify the order in which the player plays the card.

[0038] The card rank information acquisition unit is used to accurately obtain the current game's rank (one of 2, 3, 4, 5, 6, 7, 8, 9, 10, J, Q, K, or A) by performing regular expression matching and semantic analysis on the card descriptions in the text data. Combining historical card play data with the game rules, it determines the importance and role of the rank in the current situation, providing a key basis for subsequent decision-making.

[0039] The decision-making unit, which supports card play data, retrieves the card play data corresponding to the current player ID and uses association rule mining algorithms to analyze the correlation between this card play data and other hand-playing data. This unit uncovers potential connections between card play data and player strategies at specific hand stages, providing strong support for the agent's decision-making.

[0040] The comprehensive hand information management unit is used to store and manage player hand information using a hash table within a data structure, facilitating quick query and update. It also categorizes and counts hand cards by suit, rank, and other factors, analyzing their distribution to inform subsequent hand combinations and strategy development. The key reference unit for card count is used to obtain information on the number of high-level cards in a player's hand and analyze the number of high-level cards using statistical methods and data analysis models. Combining game rules with the current situation, it assesses the impact of the number of high-level cards on decision-making, providing key insights for subsequent decisions.

[0041] The effective opponent hand inference unit is used to infer the opponent's hand based on Bayesian inference methods from game theory, based on played cards, player behavior, and historical game data. By analyzing the opponent's card playing frequency and choices at different stages of the game, the probability distribution of the opponent's hand is updated, allowing for better response strategies and improving the agent's decision-making ability.

[0042] The hand count analysis unit is used to obtain the number of cards in each player's hand and to count and analyze the opponent's hand count. Combining the game progress and the hand counts of other players, it analyzes the impact of hand count on the situation, helping the agent gain a deeper understanding of the game and develop more effective strategies.

[0043] The active play accuracy determination unit uses logical reasoning and rule-matching algorithms to determine whether the current play is the first in each round, i.e., to determine active plays. By learning from playing rules and historical playing data, the model can accurately determine active plays, providing a basis for playing strategy analysis and helping the agent better grasp the timing of playing cards. It also helps the agent better determine the current game situation and generate more accurate prompts.

[0044] The suppressed card analysis and strategy development unit is used to: obtain suppressed cards in non-active play situations and analyze them. Based on the current situation and the opponent's playing patterns, it formulates a reasonable playing strategy to improve the agent's ability to respond in passive play situations. It also more accurately adds suppressed card information to the generated prompts and provides necessary information for the subsequent playing combination unit.

[0045] The historical card play record building unit is used to obtain the cards played by the other three players in the current round and, using data storage and management techniques, to build a complete historical record of card play information. Using time series data storage, it records information such as the time of play, player, and card type. This provides historical data support for subsequent analysis and decision-making, helping the agent to summarize experience and optimize strategy.

[0046] The core support unit for acquiring card combinations is used to obtain the player's chosen card combinations and analyze and identify them through data analysis and pattern recognition. Combining the game rules with the current situation, it evaluates the rationality and effectiveness of these combinations, providing core data support for the agent's decision-making and helping it select the optimal playing strategy.

[0047] The historical card play record maintenance unit is used to comprehensively record every card played by each player at every step during the game. Using data update and maintenance algorithms, it continuously maintains historical card play records. This ensures data integrity and accuracy, and provides timely updates of the latest card play information, providing reliable data support for the agent's real-time decision-making.

[0048] The data format conversion processing unit is used to ultimately integrate the above information into JSON format. Using data conversion tools and techniques, data in different formats is converted into a unified JSON format. JSON is highly readable and extensible, facilitating subsequent data processing and analysis, and seamlessly integrating with various data analysis tools and algorithms.

[0049] RGB image feature analysis and extraction submodule: When using the RGB image of the card playing screen as the carrier, advanced computer vision technology is used to capture additional information positioning to obtain effective hand and card playing information. This includes: The RGB image to 3D point cloud mapping unit uses a deep learning-based monocular depth estimation method and a convolutional neural network-based DenseDepth model to input RGB images into a 3D point cloud mapping system. Through deep pixel analysis and spatial transformation, each pixel in the current player's hand image is converted into a set of points with 3D coordinates, providing richer spatial information for subsequent analysis. Furthermore, combined with multi-view image fusion technology, a more complete 3D point cloud model is generated from multiple RGB images at different angles, improving the accuracy and comprehensiveness of the information.

[0050] The object instance mask segmentation unit imports features such as the shape and size of each hand to accurately segment and identify objects in the 3D point cloud. The attention mechanism enhances the model's focus on key areas, improving the accuracy of hand segmentation and assigning a unique identifier to each instance, enabling precise recognition and classification of hands.

[0051] The card and hand feature extraction and comparison unit is used to extract and compare card and hand features based on the extracted 3D coordinate point set and instance mask segmentation results. This unit uses a deep learning-based feature matching algorithm to extract and compare card and hand features. By calculating the cosine similarity between feature vectors, it achieves precise matching and feature extraction between the card and hand, ensuring accurate acquisition of hand information.

[0052] The streaming data generation and subsequent processing unit is responsible for generating streaming data based on segmentation and recognition results, and then performing subsequent operations based on the streaming data processing method. This unit converts image information into a streaming data format suitable for data analysis and processing, achieving seamless integration and unified processing of image information and streaming data, providing comprehensive data support for the agent's decision-making.

[0053] 2. Agent-observable environment state feature extraction module, including: The playing combination generation unit is responsible for accurately determining all possible playing combinations for the current game, based on the playing rules of Pai Gow and the game status information obtained in the first step. By establishing a formal model of the playing rules and converting them into a computer-executable algorithm, the generated playing combinations are ensured to be accurate. Furthermore, the generated playing combinations are screened and optimized based on the current hand and information about played cards, improving their rationality and effectiveness.

[0054] The situation analysis unit is responsible for: Based on the rules of the Pai Gow game and the previously acquired game status information, it fully considers factors such as the current hand, the number of cards played, and the number of cards in the hands of both the enemy and the friend. By building a situation analysis model, it generates all possible combinations of cards that comply with the rules. This situation analysis process not only considers the current state of the game but also predicts future trends and evaluates the impact of different combinations of cards on the situation, providing a comprehensive reference for the agent to select the optimal strategy. Through this step, the agent can fully understand the available card moves in the current game, the number of cards in the hands of both the enemy and the friend, and the situation of the cards already played, providing key environmental state characteristics for the agent's decision-making.

[0055] 3. Memory storage and update mechanism module, including: The game progress recording unit is used to record each player's card moves in real time using time-series data storage and management techniques. An index structure is established for the game timeline, completely preserving the entire game's card play history. This timeline index enables the agent to quickly query and analyze game progress, better understanding the game process and preventing information omissions. Furthermore, data compression and storage optimization techniques are used to reduce data storage volume and improve storage efficiency.

[0056] The player status tracking unit is used to closely track the current player's status changes, including changes in hand size and playing strategy. By building a player status model, it monitors the player's status characteristics in real time and obtains player operation data and status information in real time.

[0057] 4. Calculation and coding sequence module for the legality of possible card plays: Based on the extracted observable state features, including hand information, played cards, hand counts for both friend and foe, card level information, and various game situation analysis results, all possible card plays for the current round are calculated. This calculation strictly adheres to the rules of Pai Dan (Chinese card game), utilizing combinatorial mathematics and logical reasoning algorithms to comprehensively and accurately generate all possible combinations that meet the rules.

[0058] Subsequently, each possible encoding of a card play is serialized into a 15-dimensional array. During the encoding process, the meaning represented by each dimension in the array is carefully designed to ensure that the key information of the card play combination can be fully and accurately expressed. The first four dimensions can respectively represent the number of cards of different suits in the card play, the next five dimensions represent the number of cards of different points, and three dimensions are used to represent the identification information of special card types (such as bombs, straights, flushes, etc.). The remaining three dimensions represent information such as the order of playing cards, the priority of playing cards, and the importance associated with the current situation. Through a carefully designed encoding method, the complex possibilities of playing cards are converted into a numerical form that is easy for computers to process, providing an efficient data representation for subsequent intelligent agent decision-making and text prompt word generation, facilitating the model to perform fast and accurate analysis and processing.

[0059] 5. Generate text prompt module The multi-source information integration unit is used to deeply fuse RGB images and streaming data, leveraging information fusion and knowledge graph technologies to enable the agent to fully access information about the game environment. By constructing a knowledge graph that integrates multi-source information, different types of information are linked, enabling deep mining and analysis. During the knowledge graph construction process, semantic annotation and relationship extraction techniques are employed to integrate image information, text information, and game rules into an organic knowledge network. This provides comprehensive knowledge support for the agent's decision-making, significantly improving its ability to understand and analyze the situation.

[0060] The text prompt generation unit is responsible for generating accurate commentary prompts by combining Pai Dan game techniques and rich historical data, using natural language generation technology and deep learning models. By establishing a knowledge base of game techniques and a case library of historical data, this knowledge and data are integrated into the text generation model. The unit learns and understands the input game information and game techniques, generating commentary prompts that are contextually appropriate and appropriate. Furthermore, through manual annotation and feedback mechanisms, the model's generation results are continuously optimized, improving the accuracy and practicality of the commentary prompts.

[0061] In the above technical solution, the beneficial effects include: Highly personalized: Commentary prompts are dynamically generated based on user preferences and context. Through in-depth analysis and learning of user data, the system meets the personalized needs of different users in different scenarios, significantly improving the user experience. Whether you are a novice or a veteran player, you can get commentary prompts that meet your needs, increasing game fun and engagement.

[0062] Strong Flexibility: The flexible system architecture can quickly adapt to different application scenarios and changing requirements, supporting a variety of input and output formats. This leverages its strengths in real-time live commentary, game tutorials, and other related fields. Furthermore, the system can be rapidly adjusted and optimized based on new requirements and changes, demonstrating broad applicability and scalability.

[0063] Cost reduction: Automation and intelligent algorithms reduce reliance on rules and templates, lowering system maintenance and update costs. Without extensive human intervention and manual adjustments, the system automatically learns and optimizes, improving system sustainability and stability. This also reduces the development and maintenance costs of fixed rules and templates, improving resource efficiency.

[0064] Improved Efficiency: The automated generation mechanism can quickly generate high-quality commentary prompts. Leveraging efficient algorithms and powerful computing resources, it reduces manual intervention and significantly improves production efficiency. In scenarios requiring rapid production of large quantities of commentary prompts, such as live broadcast commentary and large-scale content generation, this mechanism meets business requirements and provides strong support for the development of related fields.

[0065] Wide Application: This solution is not only applicable to Pai Gow (Pai Gow), but can also be extended to other similar card games, such as Dou Di Zhu (Landlord) and Mahjong. By learning and adapting to different game rules and characteristics, it can provide high-quality commentary for various card games, possessing broad application prospects and market value. It also provides new ideas and methods for information generation and processing in other related fields.

[0066] exist Figure 2 In the embodiment of the present application, a method for generating explanation prompt words is provided, comprising the following steps: The environment state observation and feature extraction module is used to collect and extract features from the original observable environment to obtain a basic data set; Utilizing the agent-observable environment state feature extraction module to generate a card-playing combination based on the basic data set, and analyzing the current situation to obtain an environment state feature set; The text prompt generation module is used to integrate multi-source information and generate commentary prompts.

[0067] In a specific embodiment, it also includes: The memory storage and update mechanism module is used to establish an index structure for the game timeline, completely preserving the entire game's card-playing records; and by establishing a player state model, the player's own state characteristics are monitored in real time; the player's operation data and state information are obtained in real time; the card-playing action possibility legality calculation and coding serialization module is used to calculate all possible card plays in the current round; all card-playing combinations that comply with the rules are generated; and each possible card-playing combination is coded and serialized.

[0068] In the above technical solution, an environmental state observation and feature extraction module is set up to collect and extract features from the original observable environment to obtain a basic data set; an intelligent agent observable environmental state feature extraction module is used to generate a card-playing combination based on the basic data set and analyze the current situation to obtain an environmental state feature set; a text prompt generation module is used to integrate multi-source information and generate commentary prompts; through an automated generation mechanism, efficient algorithms and powerful computing resources are utilized to quickly generate high-quality commentary prompts; reducing manual intervention and greatly improving production efficiency.

[0069] Specifically, the method for generating the commentary prompt words includes the following steps: Step 1: Given the current hand and historical card play information, as well as the current card play situation, and the agent represented by a large language model, the agent's visual environment state observations, including RGB images, text information, and other multi-source information, are combined. If the chess game screen is used as the input, the input is a first-person RGB image of each player's card play. If the agent's real-time state observation data is represented by streaming data, such as JSON or XML text format, the input is a collection of information representations of the game environment at each moment, including serial data for all players, recording information such as their hand.

[0070] Step 2: Based on the observation of the game environment state, extract the observable environment state features of the agent, parse and extract the observable environment state features of the agent, including the optional card-playing actions under the current game conditions, the number of cards in the hands of both the enemy and the friend, the information of the cards that have been played, and the card information of the current game.

[0071] Step 3: Based on the extracted observable state features, a unique memory storage and update mechanism is constructed through the large language model to simulate the large language model-based agent's real-time understanding and memory of the current game state. This large language model memory storage and update mechanism, used to simulate the agent's understanding and memory of the current game state, is dynamic, adaptive, and continuously updated, simulating the large language model-based agent's understanding and memory of the game state.

[0072] Step 4: Based on the extracted observable state features, calculate all possible card plays in the current round and serialize each possible card play code into a 15-dimensional array.

[0073] Step 5: Combining the techniques and rules of Pai Dan (Pai Dan) with analysis of historical card play and the current situation, we select the optimal action from the available action space and develop a textual prompt to explain the action. The agent, represented by the large language model, makes a decision and selects an action from the available options. To design prompts for the large language model to simulate the agent's decision-making reasoning, we define a function to convert JSON data into natural language text data. The step function is as follows: defgenerate_commentary(id,handcard,actions,history_data): game_techniques="There are many different techniques for playing Pai Dan. Memorizing cards is key; clearly remembering the cards you've played helps you better grasp the situation. In the tribute-giving and return-giving phases, clever decisions can establish an advantage. Bombs are powerful, and their proper use can control the tempo. Heart combinations are versatile and can create powerful hand combinations. When playing with double tributes, accurately assess the situation and grasp the timing of your plays. Card control is essential for advanced players, allowing them to effectively control the situation. Sending cards requires strategy to help teammates play. Drawing cards requires timing to obtain key cards. Inserting and sorting cards keep your hand organized for easier combination. Combining fire requires skillful coordination to maximize its power. In the final stages of the game, decisively attack to secure victory." basic_prompt=f"Your current seat number is {id}, the cards in your hand are {handcard}, and your historical card play information is {history_data}. " full_prompt=( f"{basic_prompt}Analyze the current situation in depth based on {game_techniques}." From the perspective of professional Pai Dan commentary and teaching, combining Pai Dan techniques, historical play records, and the cards in your hand, carefully analyze the playing psychology of the previous and next players. From the available actions {actions} in the current round, choose the action you think is the best and analyze the reasons for making that action. f"Including the impact on the game trend, how to cooperate with teammates, and how to deal with opponents' strategies." ) returnfull_prompt.

[0074] In the above technical solution, the method analyzes the hand information of the current card game, historical card playing information, optional action combinations, and the cards played in the current game state as input, and uses a large language model to deeply understand and memorize the current card game. The large language model conducts a comprehensive analysis based on the input prompt words, and uses game theory and decision analysis methods to select the optimal action from the optional actions. At the same time, the large language model compares the difference between the decision action and the player's selected action, combines game skills and historical data, outputs information about the previous and next players' cards played in the current game, and generates commentary prompt words. Through the method of this application, it is possible to generate personalized, high-quality commentary prompt words to meet the needs of different users and improve the quality and efficiency of game commentary.

[0075] Those skilled in the art will appreciate that the present application may be implemented as a system, method, or computer program product.

[0076] Therefore, the present disclosure may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the present disclosure may be implemented in the form of a computer program product embodied in one or more computer-readable media, wherein the computer-readable media contains computer-readable program code.

[0077] Any combination of one or more computer-readable media can be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or device.

[0078] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application. Various substitutions and improvements may be made to the present application on this basis, all of which fall within the scope of protection of the present application.

Claims

1. A system for generating commentary prompt words, characterized in that: include: The environmental state observation and feature extraction module is used to collect and extract features from the original observable environment information to obtain the basic data set; The agent can observe the environment state feature extraction module, which is used to generate a card combination based on the basic data set and analyze the current situation to obtain the environment state feature set; Generate text prompt module, which is used to integrate multi-source information and generate explanation prompt words.

2. The system for generating commentary prompt words according to claim 1, wherein: Also includes: The memory storage and update mechanism module is used to establish an index structure for the game timeline and fully preserve the card play records of the entire game; and By establishing a player status model, the player's status characteristics can be monitored in real time; Get players' operation data and status information in real time.

3. The system for generating commentary prompt words according to claim 2, wherein: Also includes: The module for calculating the legality of possible card plays and encoding serialization is used to calculate all possible card plays in the current round. Generate all card combinations that meet the rules; And each card played may be encoded and serialized.

4. The system for generating commentary prompt words according to claim 3, wherein: The environmental state observation and feature extraction module includes: Multi-source information combination submodule, used to integrate multi-source information including RGB images and text information; The streaming data feature parsing and extraction submodule is used to assist the intelligent agent in aligning the information of the original observable environment with the real-time status observation data of the intelligent agent represented in text format through the adaptive template generation technology based on deep learning; the RGB image feature parsing and extraction submodule is used to use computer vision technology to locate and capture information when using the RGB image of the card-playing screen as the carrier to obtain the hand and card-playing information.

5. The system for generating commentary prompt words according to claim 4, characterized in that: The agent observable environment state feature extraction module includes: The card playing combination generation unit is used to generate all optional card playing combinations in this card playing by using rule reasoning and algorithm generation technology.

6. The system for generating commentary prompt words according to claim 5, characterized in that: The memory storage and update mechanism module includes: The game progress recording unit is used to establish an index structure for the game timeline and fully save the card play records of the entire game; The player status tracking unit is used to monitor the player's own status characteristics in real time by establishing a player status model; and to obtain the player's operation data and status information in real time.

7. The system for generating commentary prompt words according to claim 6, wherein: The card playing probability calculation and coding serialization module includes: The card playing possibility calculation unit is used to calculate all possible card playing possibilities in the current round and generate all card playing combinations that meet the rules; The coding serialization unit is used to code and serialize each possible card played.

8. The system for generating commentary prompt words according to claim 7, wherein: The text prompt generation module includes: Multi-source information integration unit, used to deeply fuse RGB images and streaming data to obtain comprehensive environmental information; The text prompt generation unit is used to generate commentary prompt words based on the comprehensive environmental information using natural language generation technology and deep learning models.

9. A method for generating commentary prompt words, characterized in that: The following steps are involved: The environment state observation and feature extraction module is used to collect and extract features from the original observable environment to obtain a basic data set; Utilizing the agent-observable environment state feature extraction module to generate a card-playing combination based on the basic data set, and analyzing the current situation to obtain an environment state feature set; The text prompt generation module is used to integrate multi-source information and generate commentary prompts.

10. The system for generating commentary prompt words according to claim 9, characterized in that: Also includes: Use the memory storage and update mechanism module to establish an index structure for the game timeline, and fully save the entire game's card play records; And by establishing a player status model, the player's status characteristics can be monitored in real time; Acquire the player's operation data and status information in real time; use the card-playing action possibility legality calculation and coding serialization module to calculate all possible card plays in the current round; generate all card-playing combinations that comply with the rules; and code and serialize each possible card-playing action.

Images