CN112138409A

CN112138409A - Game result prediction method, device and storage medium

Info

Publication number: CN112138409A
Application number: CN202010927955.7A
Authority: CN
Inventors: 杨泽龙; 王琰; 刘晓江
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-09-07
Filing date: 2020-09-07
Publication date: 2020-12-29
Anticipated expiration: 2040-09-07
Also published as: CN112138409B

Abstract

The application relates to a method, a device and a storage medium for predicting game results, wherein the method for predicting the game results comprises the following steps: determining a plurality of game characteristic data sets of each of at least two competitors from the game image frames at the predicted time point, wherein different game characteristic data sets belong to different characteristic types; determining the independent forecasting win rate value of each competitor according to all game characteristic data groups of the same characteristic type, wherein different characteristic types of the same competitor correspond to one independent forecasting win rate value respectively; determining a transient weight value corresponding to each feature type according to the predicted time point; and generating a game result according to all the independent prediction win rate values and the instant weight values, wherein the game result comprises the comprehensive prediction win rate value of each competitor and the characteristic contribution value corresponding to each characteristic type, so that accurate and interpretable real-time win rate prediction information can be provided when the game result is predicted, and the game prediction content is enriched.

Description

Game result prediction method, device and storage medium

Technical Field

The present application relates to the field of game technologies, and in particular, to a method and an apparatus for predicting a game result, and a storage medium.

Background

In the current internet digital age, electronic games have become an important part of national life, and MOBA (Multiplayer Online Battle Arena) games are the most popular type of electronic games at present. Nowadays, MOBA electronic competitions become regular items in sports competition such as the sub-fortune meeting, and activities such as live broadcast and offline competition of various games create huge economic wealth for the society, and greatly enrich the lives of people. Under the background of the times, the MOBA game winning rate prediction system can be applied to various aspects such as formation recommendation, live competition, virtual anchor talk generation and the like, and has wide application prospect and great economic value.

However, when the game result is predicted by the conventional MOBA game winning rate prediction system, the obtained game result includes only winning rates of both parties of the battle in the game, and the contents are single, and the practical value is low.

Disclosure of Invention

The embodiment of the application provides a method, a device and a storage medium for predicting a game result, which are used for solving the problems that the obtained game result is single in content and low in practical value because only the success rate predicted values of two parties of a match in a game can be given in the process of predicting the game result.

The embodiment of the application provides a method for predicting a game result, which comprises the following steps:

determining a plurality of game feature data sets of each of at least two competitors from the game image frames at the predicted time point, wherein each game feature data set comprises at least one game feature data, and different game feature data sets belong to different feature types;

determining the independent forecasting win rate value of each competitor according to all game characteristic data groups of the same characteristic type, wherein different characteristic types of the same competitor correspond to one independent forecasting win rate value respectively;

determining a transient weight value corresponding to each feature type according to the predicted time point;

and generating a game result according to all the independent prediction win value and the instantaneous weight value, wherein the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type.

The embodiment of the present application further provides a device for predicting a game result, including:

the first determining module is used for determining a plurality of game characteristic data sets of each of at least two competitors from the game image frames at the predicted time point, each game characteristic data set comprises at least one game characteristic data, and different game characteristic data sets belong to different characteristic types;

the second determining module is used for determining the independent prediction win rate value of each competitor according to all game feature data groups of the same feature type, and different feature types of the same competitor correspond to one independent prediction win rate value respectively;

the third determining module is used for determining the instant weight value corresponding to each feature type according to the predicted time point;

and the first generation module is used for generating a game result according to all the independent prediction win value and the instant weight value, wherein the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type.

Wherein, the second determining module specifically comprises:

the determining unit is used for determining a win rate prediction submodel corresponding to each characteristic type from the trained win rate prediction models;

the processing unit is used for processing the game feature data group of the corresponding feature type by using the win rate forecasting submodel to obtain an independent forecasting win rate value of each competitor of the corresponding feature type;

the third determining module is specifically configured to:

and determining the instantaneous weight value corresponding to each feature type by using the time weight submodel and the prediction time point in the trained win rate prediction model.

Wherein, the prediction device of the game result further comprises:

the acquisition module is used for acquiring a plurality of game videos which are finished, each game video comprises at least two historical competitors, and the final win ratio value of each historical competitor in each game video is acquired;

the extraction module is used for extracting a plurality of game sample image frames at selected time points from each game video;

a fourth determining module, configured to determine, from each game sample image frame, a plurality of game sample feature data sets of each historical competitor, where each game sample feature data set includes at least one game sample feature data, and different game sample feature data sets belong to different feature types;

and the training module is used for training the preset win rate prediction model according to the selected time point, the feature type, the final win rate value and the game sample feature data set so as to obtain the trained win rate prediction model.

The preset win rate prediction model comprises a plurality of win rate prediction submodels and a time weight submodel, different feature types respectively correspond to one win rate prediction submodel, and the training module specifically comprises:

the first training unit is used for training the corresponding winning rate prediction submodel according to all game sample feature data groups of the same feature type and the final winning rate value;

and the second training unit is used for training the time weight submodel according to the game sample feature data sets of all the feature types at each selected time point and the final win value.

The characteristic types comprise an economic characteristic type, a killing characteristic type and a defense tower characteristic type, the multiple winning rate prediction submodels comprise a first logistic regression model, a second logistic regression model and a third logistic regression model, and the first training unit is specifically used for:

training the first logistic regression model according to all game sample feature data sets of economic feature types and the final win rate value;

training the second logistic regression model according to all game sample feature data groups of the killing feature types and the final win probability value;

and training the third logistic regression model according to all game sample feature data sets of the defense tower feature types and the final win rate value.

The characteristic types comprise wild area resource characteristic types, soldier line characteristic types and formation characteristic types, the multiple winning rate prediction submodels comprise a first neural network model, a second neural network model and a third neural network model, and the first training unit is specifically used for:

training the first neural network model according to all game sample characteristic data groups of the wild area resource characteristic types and the final win probability value;

training a second neural network model according to all game sample characteristic data sets of the solidus characteristic types and the final win probability value;

and training the third neural network model according to all game sample feature data sets of the formation feature types and the final win rate value.

Wherein, the prediction device of the game result further comprises:

the second generation module is used for generating a corresponding comment text according to the prediction time point, the comprehensive prediction win ratio value and the characteristic contribution value;

and the providing module is used for providing the comment text for the user in a preset output mode.

The embodiment of the application also provides a computer readable storage medium, wherein a plurality of instructions are stored in the storage medium, and the instructions are suitable for being loaded by a processor to execute any game result prediction method.

The embodiment of the present application further provides a server, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in any one of the methods for predicting a game result when executing the computer program.

The application provides a method, a device and a storage medium for predicting game results, wherein a plurality of game characteristic data sets of each of at least two competitors are determined from a game image frame at a prediction time point, each game characteristic data set comprises at least one game characteristic data, different game characteristic data sets belong to different characteristic types, an independent prediction win value of each competitor is determined according to all game characteristic data sets of the same characteristic type, different characteristic types of the same competitor correspond to one independent prediction win value respectively, then an instant weight value corresponding to each characteristic type is determined according to the prediction time point, a game result is generated according to all the independent prediction win values and the instant weight values, the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type, therefore, when the game result is predicted, the win rate value of each competitor in the game can be given, and the prediction basis of the win rate value can be given, so that accurate and interpretable real-time win rate prediction information can be provided, the game prediction content is enriched, and the practical value is improved.

Drawings

The technical solution and other advantages of the present application will become apparent from the detailed description of the embodiments of the present application with reference to the accompanying drawings.

FIG. 1 is a schematic diagram of a scenario of a system for predicting game results according to an embodiment of the present disclosure;

FIG. 2 is a flow chart illustrating a method for predicting game results according to an embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a process for predicting game results provided by an embodiment of the present application;

FIG. 4 is a schematic flow chart of a method for predicting game results according to an embodiment of the present disclosure;

FIG. 5 is a schematic diagram of the temporal weight values of six feature types as a function of game time provided by an embodiment of the present application;

FIG. 6 is a schematic flow chart of a method for predicting game results according to an embodiment of the present disclosure;

FIG. 7 is a schematic diagram of the composite predicted win value of two competitors according to the embodiment of the present application;

8-10 are diagrams of an illustrative interface for different game time points provided by the embodiment of the application;

FIG. 11 is a schematic structural diagram of a device for predicting game results according to an embodiment of the present application;

fig. 12 is a schematic structural diagram of a server according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Artificial Intelligence (AI) is a comprehensive subject, and relates to a wide range of fields, both hardware and software technologies. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

Machine Learning (ML) is a multi-domain cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for making computers have intelligence, and its application is spread in each field of artificial intelligence, for example, can be applied to online Games, such as in Multiplayer online tactics sports Games (MOBA), and can be used as the appurtenance of fighting and competition, make the user when the fighting or live broadcast of competition, can utilize recreation AI to predict the game victory or defeat, and then not only can make things convenient for novice to better understand the situation, can also improve the interest of fighting.

The scheme provided by the embodiment of the application relates to an artificial intelligence machine learning technology, in particular to a method and a device for predicting a game result and a storage medium.

Referring to fig. 1, fig. 1 is a schematic view of a scene of a game result prediction system according to an embodiment of the present disclosure, where the game result prediction system may include any one of the game result prediction devices according to the embodiments of the present disclosure, and the game result prediction device may be specifically integrated in a server, such as a game server, where the server may be a single server or a server cluster composed of multiple servers.

The server may determine, from the game image frames at the predicted time points, a plurality of game feature data sets for each of at least two competitors, each game feature data set including at least one game feature data, different game feature data sets belonging to different feature types; determining the independent forecasting win rate value of each competitor according to all game characteristic data groups of the same characteristic type, wherein different characteristic types of the same competitor correspond to one independent forecasting win rate value respectively; determining a transient weight value corresponding to each feature type according to the predicted time point; and generating a game result according to all the independent prediction win value and the instantaneous weight value, wherein the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type.

In addition, the system for predicting the game result may further include a terminal installed with a game application, where the terminal may be a smart phone, a tablet Computer, an intelligent bluetooth device, a notebook Computer, or a Personal Computer (PC), and the terminal may be connected to the server through a network, and may receive the game result sent by the server, so as to provide the game result to the user. Specifically, the terminal may be a device running a game, and the game result may be a result obtained by predicting whether a server performs a game in which the terminal is currently playing at a preset prediction time point, where the game may be a multiplayer online tactical competitive game, a chess-card game, a role-playing game, or the like, each competitor in the game includes at least one player, and different competitors have a confrontation relationship.

For example, as shown in fig. 1, when the current game progresses to a predicted time point, for example, 1 minute of play, the server may acquire a game image frame at 1 minute of play of the game, extract game feature data groups a1, a2, A3, a4, a5, and a6 corresponding to six feature types 1, 2, 3, 4, 5, and 6 of a competitor a, and game feature data groups B1, and B1 corresponding to six feature types 1, 2, 3, 4, 5, and 6 of a competitor B from the game image frame, and then determine corresponding feature data values a1 and B1, and predict corresponding feature values of different types of a, a1, B1, a1, and B1 of the competitor a, and predict corresponding feature values of different types of the respective competitors a1, and B1, B, B and B, then determining the instantaneous weight values T, T and T of six feature types 1 to 6 at the prediction time point, then taking the product of the independent prediction probability value and the instantaneous weight value of each feature type of the competitor A as the feature contribution value of the feature type of the competitor A, taking the product of the independent prediction probability value and the instantaneous weight value of each feature type of the competitor B as the feature contribution value of the feature type of the competitor B, taking the sum of the feature contribution values of the six feature types of the competitor A as the comprehensive prediction probability value of the competitor A, namely (A + T + A + T), and taking the sum of the feature contribution values of the six feature types of the competitor B as the comprehensive prediction probability value of the competitor B, namely (B + T + B + T), to obtain the game result of the current game at the predicted time point.

As shown in fig. 2, fig. 2 is a schematic flow chart of a game result prediction method provided in an embodiment of the present application, and a specific flow of the game result prediction method may be as follows:

s101, determining a plurality of game characteristic data sets of each of at least two competitors from game image frames at the predicted time point, wherein each game characteristic data set comprises at least one game characteristic data, and different game characteristic data sets belong to different characteristic types.

In this embodiment, the method for predicting the game result may be applied to different game applications, for example, a PvP (Player VS Player) playing method in a game such as a multiplayer online Battle game (MOBA), a chess game, a role playing game, and the like.

For convenience of description and understanding, the embodiment of the present application is illustrated by taking a MOBA game as an example. In an MOBA network game scene, players are generally divided into two teams (i.e., two opponents), the two opponents compete with each other in a dispersed game map, in the map, besides virtual hero characters selected by the two players, Non-Player controlled game units (Non-Player Character, NPC) such as soldiers, defense towers, small wild monsters, special wild monsters and the like are also arranged, each Player can control the selected virtual hero characters to hit enemy hero or middle cubic units on the map to obtain resources, and finally destroy an enemy base to obtain a final win. In the game, the game fighting data of both players constantly changes, and by acquiring the game characteristic data at the predicted time point in the game fighting data of both players, the winning rate of the participating players can be dynamically predicted.

Specifically, the predicted time point may be a time point selected by the user and required to perform the game result prediction, or may be a time point required to perform the game result prediction preset by the system, for example, the game result prediction may be performed every preset time interval from the game starting time, and a plurality of corresponding predicted time points may be provided, for example, the preset time interval is 30 seconds, and the plurality of predicted time points may be sequentially 0 second for starting, 30 seconds for starting, 60 seconds for starting, etc., until the game is ended. The plurality of game feature data sets may include game feature data that contributes to the prediction of the game result, such as the accumulated virtual money of the game opponent at the predicted time point, the number of selected enemy heroes, the number of defense towers, mastered strange information (e.g., the number of violent men, the number of dominant men, or the number of dark violent men), soldier-to-enemy crystal distance information (e.g., the distance from an on-road soldier to an enemy crystal, or the distance from an off-road soldier to an enemy crystal), and virtual hero formation information.

The game feature data sets of different feature types can be understood as prediction bases for predicting game results from different angles, and in specific implementation, the feature types and their dependencies with the game feature data can be predefined artificially, for example, determined by studying analysis ideas and judgment bases adopted in game result prediction by real human beings.

In one embodiment, the feature types may include economic feature types, whereas the virtual money in the game may be used to purchase more powerful equipment to enhance various attributes of the virtual hero (such as blood volume, offensive power, and defensive power) to increase the winning of the player, the accumulated virtual money at the predicted point in time by the player may belong to economic feature types; in addition, in view of the fact that each time the own virtual hero kills one enemy hero in the game, extra money and experience can be obtained for purchasing equipment and upgrading the hero, so that the winning of the own party is increased, the characteristic types can also comprise killing characteristic types, and the number of the enemy hero killed by the game competitor at the predicted time point can belong to the killing characteristic types; in addition, the defense towers in the game can resist local soldiers and heroes and can provide visual fields, and the game has great significance for competition victory and defeat, the characteristic types can also comprise defense tower characteristic types, and the number of the defense towers of the game competitors at the predicted time point can belong to the defense tower characteristic types; in addition, in view of the fact that the team killing the wild monster in the game can obtain additional gains including money, experience and attribute enhancement, the characteristic types can also comprise wild area resource characteristic types, and the wild monster information mastered by the game opponent at the predicted time point can belong to the wild area resource characteristic types; in addition, as the soldiers in the game can be used for destroying enemy defense towers, the characteristic types can also comprise a soldier line characteristic type, and the distance information from the soldiers to the enemy crystal at the predicted time point of the game competitor can belong to the soldier line characteristic type; in addition, in view of the fact that different virtual heroes in the game have different attributes and skills and the coordination and control relationship exists between the virtual heroes and the virtual heroes, the characteristic types can also comprise formation characteristic types, and the formation information of the virtual heroes selected by the game competitor at the predicted time point can belong to the formation characteristic types. Therefore, the game winning rate can be independently predicted from a plurality of angles such as economy, killing, defense towers, wild area resources, soldier lines, formation and the like in the subsequent steps, and accurate and interpretable fact winning rate prediction information can be provided.

In specific implementation, as shown in the upper left part of fig. 3, when the current game progresses to a predicted time point (for example, 1 minute of play), the server may obtain a game image frame of the predicted time point of the game, extract a plurality of game feature data belonging to different feature types (for example, economy, killing, defense tower, field resource, soldier line and formation) and time features of each of all the competitors from the game image frame, and group the game feature data of the same feature type of each competitor into a group to obtain a plurality of game feature data groups of different feature types of each competitor.

S102, determining the independent forecasting win ratio value of each competitor according to all game characteristic data groups of the same characteristic type, wherein different characteristic types of the same competitor correspond to one independent forecasting win ratio value respectively.

In this embodiment, the independently predicted win value for each feature type of the player in the game is only associated with all game feature data sets for that feature type, regardless of the length of time that the game has been played. Taking the economic characteristic type as an example, in the MOBA game, all game characteristic data sets of the economic characteristic type may include virtual money accumulated by two competitors at a predicted time point, and the probability that the party accumulating more virtual money wins is greater.

As shown in fig. 4, the S102 may specifically include:

and S1021, determining a winning rate prediction submodel corresponding to each feature type from the trained winning rate prediction model.

Specifically, as shown in the upper right part of fig. 3, the trained win rate prediction model may include a plurality of win rate prediction submodels, and the win rate prediction submodels correspond to the feature types (e.g., economy, attack, defense tower, wild resources, soldier lines or formation) in a one-to-one manner, that is, one win rate prediction submodel corresponds to each feature type, so as to individually predict the win rate of each party in the game based on the game feature data of each feature type. The victory ratio prediction submodel can be a machine learning model such as a logistic regression model, a feedforward neural network model, a long-short term memory network model or a gated cyclic unit network model.

And S1022, processing the game feature data group of the corresponding feature type by using the win rate prediction submodel to obtain the independent prediction win rate value of each competitor of the corresponding feature type.

Specifically, all game feature data sets of the same feature type may be used as inputs to the corresponding win rate prediction submodel, and the win rate prediction submodel may be caused to output an independent win rate prediction value for each competitor of the corresponding feature type. Taking the economic characteristic type as an example, in the MOBA game, all game characteristic data sets of the economic characteristic type may include virtual money accumulated by two competitors at the predicted time point, for example, if the competitor a has the virtual money 5300 and the competitor B has the virtual money 4800, then the { competitor a-economic characteristic type: 5300, Containment B-economic characteristics type: 4800 as the input of the winner rate predictor model corresponding to the economic feature type to obtain the independent winner rate predictors corresponding to the economic feature types of two competitors in the MOBA game outputted by the winner rate predictor model, for example, { competitor a-economic feature type: 55.2%, competitor B-economic characteristics type: 44.8% }.

In some alternative embodiments, as shown in the upper right part of fig. 3, the feature types may be divided into two broad categories, the first category of feature types may include economic, killing or defense tower feature types, and the data types of the game feature data sets of these feature types are numerical types, the second category of feature types may include wild monster, soldier line or formation, and the data types of the game feature data sets of these feature types are category types, and S1022 may specifically include:

s1-1, all game feature data sets of the economic feature type, the killing feature type or the defense tower feature type are normalized, and the game feature data sets after the normalization processing are input into the corresponding winning rate prediction submodels, so that the independent prediction winning rate value of each competitor of the corresponding feature types is obtained.

Specifically, each game feature data group may include at least one game feature data, and in the MOBA game, all players participating in the game are divided into two teams, that is, there are two competing parties, and the number of all game feature data groups of the same feature type is two, so that the number of all game feature data of the same feature type is at least two, and the data type thereof is a numeric type. In specific implementation, the game feature data in all game feature data groups of the economic feature type, the killing feature type or the defense tower feature type can be normalized according to a normalization formula, wherein the normalization formula can be as follows:

v＝(x-vmin)/(vmax-vmin)；

v is the value of the game feature data in the game feature data group after the value x of the game feature data in the game feature data group is normalized; vmin is the minimum value of the game characteristic data in all the game characteristic data groups; vmax is the maximum value of game feature data in all game feature data sets.

Taking the economic feature type as an example, in the MOBA game, all game feature data of the economic feature type may be { competitor a-economic feature type: 5300}, { party B-economic feature type: 4800, all game feature data of the normalized economic feature type may be { competitor a-economic feature type: 1}, { party B-economic feature type: 0}.

S1-2, converting all game feature data groups of the wild monster feature type, the soldier line feature type or the formation feature type into feature vectors, and inputting the feature vectors into corresponding win rate predictor models to obtain independent prediction win rate values of each competitor of the corresponding feature types.

The dimension number of the feature vector can be determined by the number of data categories included in the game feature data set, and the dimension value can be determined by the corresponding category value. Taking the lineup feature type as an example, in the MOBA game, all game feature data belonging to the lineup feature type may include { party a-lineup feature type: trip biting; month MI; simayi; a dun mountain; marcobollo }, { party B-formation feature type: carrying out magic ox; mink cicada; shining; catching a tiger in a field; canada, each game feature data includes five hero categories, and the category value corresponding to each hero category may be an Identification (ID) number of the corresponding hero. Accordingly, the feature vectors obtained by converting all game feature data sets of the lineup feature type may be (ID11, ID12, ID13, ID14, ID15, ID21, ID22, ID23, ID24, ID25), where ID11 to ID15 are the ID numbers of the five virtual heros selected by the competitor a, and ID21 to ID25 are the ID numbers of the five virtual heros selected by the competitor B.

S103, determining the corresponding instantaneous weight value of each feature type according to the predicted time point.

In this embodiment, as shown in the lower right part of fig. 3, the instantaneous weight value of each feature type is used to represent the relative importance of the feature type at the prediction time point, and the larger the instantaneous weight value is, the larger the influence of the game feature data of the corresponding feature type on the final win or loss of the game is. Wherein the instantaneous weight values of the same feature type are time-varying, and the sum of the instantaneous weight values of all feature types is fixed, for example, 1.

In the MOBA game, as shown in fig. 5, at the time point of 0 minute of the game, that is, when the game is played, the instantaneous weight values of the other feature types may be all zero except that the instantaneous weight value of the lineup feature type is not zero, and at this time, the instantaneous weight value of the lineup feature type may be 1. As the game progresses, the instantaneous weight value of the economic feature type increases rapidly and starts to decrease at a later stage of the game, for example, 12.0 minutes after the game time point, because the extra money exceeding a certain threshold becomes useless at the later stage of the game due to the limited packaging space of the equipment. After 19.0 minutes has elapsed at the point in time of play, the instantaneous weight value for the soldier trait type may be greater than the instantaneous weight value for the economic trait type because the play trait data for the soldier trait type (e.g. the distance of the soldier from the competitor defense tower) becomes more important after 19.0 minutes at the point in time of play. The instantaneous weight value of the wild area resource feature type will gradually increase after 8.0 minutes of the game time point because the wild monsters "overlord" and "dark tyrant" will appear only after 8.0 minutes and 10.0 minutes of the game time point, respectively.

Specifically, with continued reference to fig. 4, the aforementioned S103 may include: and determining the instantaneous weight value corresponding to each feature type by using the time weight submodel and the prediction time point in the trained win rate prediction model.

The server may use the predicted time point as an input of a time weight submodel, and enable the time weight submodel to input an instantaneous weight value of each feature type at the predicted time point, wherein the instantaneous weight value of each feature type is only related to the predicted time point at which the game result prediction needs to be performed, and is not related to the game feature data belonging to the feature type. Specifically, the time weight submodel may be a machine learning model such as a logistic regression model, a feedforward neural network model, a long-short term memory network model, or a gated cyclic unit network model.

In an embodiment, before the step S1021, in order to obtain the trained win ratio prediction model, the method may further include:

and step A, obtaining a plurality of game videos which are ended, wherein each game video comprises at least two historical competitors, and obtaining the final win rate value of each historical competitor in each game video.

Specifically, the server may obtain game videos with a plurality of completed game videos (e.g., 10000 game videos), and the game videos may be generated by high-skill human players (e.g., the top 1% of the ranking list of the game) for competition, so as to reduce random factors affecting the final win value in the game, thereby facilitating the reliability of a win prediction model established based on the obtained game videos.

And B, extracting a plurality of game sample image frames at selected time points from each game video.

For each game video, the selected time points may include game starting time of a game corresponding to the game video, and the time intervals between any two adjacent selected time points may be equal, for example, 30 seconds. For example, if the total duration of a game video is 15 minutes, 31 game sample image frames with game time points of 0 second, 30 seconds, 60 seconds, and 900 seconds may be extracted from the game video.

And step C, determining a plurality of game sample characteristic data sets of each historical competitor from each game sample image frame, wherein each game sample characteristic data set comprises at least one game sample characteristic data, and different game sample characteristic data sets belong to different characteristic types.

Specifically, the server may extract a plurality of game sample feature data belonging to different feature types for each of at least two competitors from each game sample image frame, and group the game sample feature data of the same feature type for each competitor to obtain a plurality of game sample feature data groups belonging to different feature types for each competitor.

And D, training a preset win rate prediction model according to the selected time point, the feature type, the final win rate value and the game sample feature data set to obtain the trained win rate prediction model.

In this embodiment, the preset win rate prediction model may include a plurality of win rate prediction submodels and a time weight submodel, and different feature types each correspond to one win rate prediction submodel, wherein, in specific implementation, Cross-Entropy Loss function (Cross-Entropy Loss) and Back Propagation (BP) algorithm may be used to train and optimize relevant parameters of each win rate prediction submodel and the time weight submodel, and the win rate prediction submodel and the time weight submodel in the win rate prediction model may also be separately trained or synchronously trained.

Specifically, when the win ratio prediction submodel and the time weight submodel in the win ratio prediction model are trained synchronously, the step D may specifically include:

and S1-1, training the corresponding winning rate prediction submodel according to the characteristic data set of all game samples of the same characteristic type and the final winning rate value.

In one embodiment, the feature types may include an economic feature type, a killing feature type, and a defense tower feature type, the plurality of win rate predictor models may include a first logistic regression model, a second logistic regression model, and a third logistic regression model, and the S1-1 may specifically include:

s1-1-1, training the first logistic regression model according to all game sample feature data sets of economic feature types and the final win rate value.

S1-1-2, training a second logistic regression model according to the characteristic data set of all game samples of the killing characteristic types and the final win probability value.

S1-1-3, training a third logistic regression model according to all game sample feature data sets of the defense tower feature types and the final win ratio value.

In another embodiment, the feature types may further include a wilderness resource feature type, a soldier line feature type, and a formation feature type, the plurality of win rate predictor models may further include a first neural network model, a second neural network model, and a third neural network model, and the S1-1 may further specifically include:

s1-1-4, training the first neural network model according to all game sample feature data sets of the wild area resource feature types and the final win rate value.

S1-1-5, training a second neural network model according to all game sample feature data sets of the solidus feature types and the final win probability value.

S1-1-6, training a third neural network model according to all game sample feature data sets of the formation feature types and the final win rate value.

In a specific implementation, the first, second and third logistic regression models may be logistic regression models having a hyperbolic tangent (tanh) function as a nonlinear function, and the first, second and third neural network models may be double-layer feedforward neural network models having sizes of (256, 16), (256, 16) and (128, 16), respectively. Also, before entering the layers, all game sample feature data sets of the respective feature types may be normalized (for the first, second, and third logistic regression models) using the minimum and maximum values, or preprocessed as embedded vectors (for the first, second, and third two-layer feedforward neural network models); after the concealment layer, the first, second, and third bi-layer feedforward neural network models described above may use a Leaky linear rectification (Leaky ReLU) function with a parameter of 0.01 as the non-linear function and a ratio of 0.2 of the layers of the conjugate to mitigate overfitting; after the output layers of the first, second, and third dual-layer feedforward neural network models described above, the outputs (winning scores) of the first, second, and third dual-layer feedforward neural network models may be corrected to values within the range of [ -1.0, 1.0] values using the tanh function to represent the likelihood of win or loss.

S1-2, training the time weight sub-model according to the game sample feature data set of all feature types of each selected time point and the final win value.

The temporal weight submodel includes a weight parameter of each of all feature types, and the weight parameter of each feature type is a time function, wherein the instantaneous weight value at the prediction time point may be understood as a weight parameter value of the corresponding feature type at the prediction time point.

It should be noted that, although the above-mentioned embodiments illustrate six feature types including an economic feature type, a killing feature type, a defense tower feature type, a wilderness resource feature type, a soldier line feature type and a formation feature type, the above-mentioned embodiments are not limited to include the six feature types, and the specific implementation may be determined according to the game application to which the above-mentioned method for predicting the game result is specifically applied, that is, the game application to which the above-mentioned method for predicting the game result is applied may be different, and the corresponding feature types may be different.

And S104, generating a game result according to all the independent prediction win value and the instantaneous weight value, wherein the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type.

Specifically, with continued reference to the lower right portion of fig. 3, for each competitor, the product of the independent predicted win value and the instantaneous weight value of each feature type of the competitor may be used as the feature contribution value of the corresponding feature type of the competitor, and the sum of the feature contribution values of all feature types of the competitor may be used as the comprehensive predicted win value of the competitor, so as to obtain the game result of the current game at the predicted time point. Thus, compared with the reason that the conventional game winning rate prediction system can only give a predicted winning rate value but cannot give the prediction, the game result prediction method in the embodiment can not only provide accurate real-time winning prediction, but also attribute the final winning rate value prediction result to the contribution of game feature data of different feature types, so that the game result prediction method can be used for commentators to refer to and make more attractive comments.

As can be seen from the above, the method for predicting game results provided by this embodiment determines a plurality of game feature data sets for each of at least two competitors from a game image frame at a prediction time point, each game feature data set includes at least one game feature data, different game feature data sets belong to different feature types, and determines an independent predicted win value for each competitor according to all game feature data sets of the same feature type, each of the different feature types of the same competitor corresponds to an independent predicted win value, then determines an instantaneous weight value corresponding to each feature type according to the prediction time point, and generates a game result according to all the independent predicted win values and the instantaneous weight values, the game result includes an integrated predicted win value for each competitor and a feature contribution value corresponding to each feature type, so that, when the game result is predicted, the method can not only provide the win rate value of each competitor in the game, but also provide the prediction basis of the win rate value, thereby providing accurate and interpretable real-time win rate prediction information, enriching the game prediction content and improving the practical value of the obtained game result.

As shown in fig. 6, fig. 6 is another schematic flow chart of a game result prediction method provided in the embodiment of the present application, and a specific flow of the game result prediction method may be as follows:

s201, determining a plurality of game characteristic data sets of each of at least two competitors from game image frames at the predicted time point, wherein each game characteristic data set comprises at least one game characteristic data, and different game characteristic data sets belong to different characteristic types.

S202, determining a winning rate prediction submodel corresponding to each feature type from the trained winning rate prediction models.

S203, the game feature data set of the corresponding feature type is processed by using the win rate forecasting sub-model, and the independent forecasting win rate value of each competitor of the corresponding feature type is obtained.

And S204, determining the instantaneous weight value corresponding to each feature type by using the time weight submodel and the prediction time point in the trained win rate prediction model.

S205, generating a game result according to all the independent prediction win value and the instant weight value, wherein the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type.

It should be noted that, the specific implementation manners of S201 to S205 in this embodiment may refer to the specific implementation manners of S101, S1021, S1022, S103, and S104 in the previous method embodiment, and therefore, the detailed description thereof is omitted here.

And S206, generating a corresponding comment text according to the prediction time point, the comprehensive prediction win ratio value and the characteristic contribution value.

Specifically, the server may generate a comment template corresponding to the predicted time point, the comprehensive success rate value, and the feature contribution value based on a preset comment text library, or may use the predicted time point, the comprehensive predicted success rate value, and the feature contribution value as inputs of a game AI commentator or a game AI commentator trained in advance to obtain a corresponding comment text.

And S207, providing the comment text for the user in a preset output mode.

The preset output mode can be a subtitle, audio or virtual host video output mode. Specifically, the S207 may include:

when the preset output mode is a subtitle, displaying an explanation text for a user;

when the preset output mode is audio, outputting voice for broadcasting the explication text;

and when the preset output mode is the video of the virtual host, displaying the virtual image, and outputting the voice matched with the virtual image to broadcast the comment text.

For a specific example, taking a live game as an example, as shown in fig. 7, fig. 7 shows a real-time comprehensive predicted win ratio value curve of two competitors a and B in the game, and then screenshots of three

time points

1, 2 and 3 in the real-time game are respectively displayed in fig. 8, 9 and 10. In the screenshot, the white chinese caption may be the real comment generated by the game AI commentator system described above. At time point 1 (tournament time 0:03), since it is the beginning of the game, the above-mentioned win ratio prediction model predicts a win probability of 54.4% for the competitor B (i.e., the blue team) based only on team composition (lineup) information, so that the game AI commentator generates a comment "start of game! Team composition of blue teams is advantageous; thus, they achieved a winning probability of 54.4%. "then, at time point 2 (race time 2:11), the probability of winning to party A (red team) drops to 36.2% due to the disadvantages of economy and killing, and the economic characteristic type and killing characteristic type are predicted to contribute to the winning characteristic as-0.187 and-0.073, respectively. Thus, the AI reviewer generates a review: the "red team is at a disadvantage in gold and lethality. AI predicts that the winning probability of the red team is only 36.2% ". Finally, at time point 3 (tournament time 9:44), party A reduced the odds of winning to 56.2%. The prediction is made mainly based on the differences in characteristic types of "economy", "wilderness resources", and "defense towers" between two teams, so AI commentators generate comments "with a truly confusing and badly beginning, and the red team gradually catches up killing more violent people and destroying more towers, thus accumulating advantages in gold. Currently, their winning odds have risen to 56.2%. "obviously, without interpretable results, the game commentator can only make some trivial commentary without attracting the audience. For example, at time point 3, only "now the twisting situation of the team, with a winning probability of 56.2%" can be reviewed, which is less interpretable information by the AI reviewer with the help of the win ratio prediction model.

As can be seen from the above, the method for predicting a game result provided in this embodiment can not only provide the win rate value of each competitor in the game, but also provide a prediction basis for the win rate value, so as to provide accurate and interpretable real-time win rate prediction information for a virtual commentary system or a human commentator, enrich the game prediction content, and improve the practical value of the obtained game result.

On the basis of the method in the foregoing embodiment, the present embodiment will be further described from the perspective of a game result prediction device, please refer to fig. 11, where fig. 11 specifically describes the game result prediction device provided in the present embodiment, which may include: a first determining module 610, a second determining module 620, a third determining module 630, and a first generating module 640, wherein:

(1) first determining module 610

The first determining module 610 is configured to determine a plurality of game feature data sets for each of at least two players from the game image frames at the predicted time points, each game feature data set including at least one game feature data, different game feature data sets belonging to different feature types.

(2) Second determination module 620

The determining module 620 is configured to determine an independent predicted win ratio value of each competitor according to all game feature data sets of the same feature type, where different feature types of the same competitor correspond to an independent predicted win ratio value respectively.

The second determining module 620 may specifically include:

and the processing unit is used for processing the game feature data group of the corresponding feature type by using the win rate forecasting submodel to obtain the independent forecasting win rate value of each competitor of the corresponding feature type.

(3) Third determination module 630

A third determining module 630, configured to determine, according to the predicted time point, an instantaneous weight value corresponding to each feature type.

In an embodiment, the third determining module 630 may be specifically configured to: and determining the instantaneous weight value corresponding to each feature type by using the time weight submodel and the prediction time point in the trained win rate prediction model.

(4) First generation module 640

A first generating module 640, configured to generate a game result according to all the independent predicted win value and the instantaneous weight value, where the game result includes a comprehensive predicted win value of each competitor and a feature contribution value corresponding to each feature type.

Wherein, in order to obtain the trained win ratio prediction model, the game result prediction device may further include:

(5) acquisition module

The acquisition module is used for acquiring a plurality of game videos which are ended, each game video comprises at least two historical competitors, and the final win rate value of each historical competitor in each game video is acquired.

(6) Extraction module

(7) fourth determining module

(8) training module

The preset win rate prediction model includes a plurality of win rate prediction submodels and a time weight submodel, and different feature types respectively correspond to one win rate prediction submodel, and the training module may specifically include:

In one embodiment, the feature types may include an economic feature type, a killing feature type, and a defense tower feature type, the plurality of win rate predictor models may include a first logistic regression model, a second logistic regression model, and a third logistic regression model, and the first training unit may be specifically configured to:

In another embodiment, the feature types may further include a wilderness resource feature type, a soldier line feature type, and a formation feature type, the plurality of win rate predictor models may include a first neural network model, a second neural network model, and a third neural network model, and the first training unit may be specifically configured to:

In the above embodiment, the prediction device of the game result may further include:

(9) second generation module

And the second generation module is used for generating a corresponding comment text according to the prediction time point, the comprehensive prediction win ratio value and the characteristic contribution value.

(10) Providing module

In specific implementation, the above units and modules may be implemented as independent entities, or may be combined arbitrarily to be implemented as the same or several entities, and specific implementations of the above units and modules may refer to the foregoing method embodiments, and are not described herein again.

As can be seen from the above, the prediction apparatus for a game result provided by this embodiment includes a first determining module, configured to determine, from a game image frame at a prediction time point, a plurality of game feature data sets for each of at least two players, where each game feature data set includes at least one game feature data, and different game feature data sets belong to different feature types; the second determining module is used for determining the independent prediction win rate value of each competitor according to all game feature data groups of the same feature type, and different feature types of the same competitor correspond to one independent prediction win rate value respectively; the third determining module is used for determining the instant weight value corresponding to each feature type according to the predicted time point; the first generation module is used for generating a game result according to all the independent prediction success rate values and the instant weight values, wherein the game result comprises the comprehensive prediction success rate value of each competitor and the characteristic contribution value corresponding to each characteristic type, so that when the game result is predicted, not only the success rate value of each competitor in the game can be given, but also the prediction basis of the success rate value can be given, accurate and interpretable real-time success rate prediction information can be further provided, the game prediction content is enriched, and the practical value is improved.

Correspondingly, an embodiment of the present application further provides a server, where the server may be a single server, or may be a server cluster composed of multiple servers, as shown in fig. 12, which shows a schematic structural diagram of a server according to an embodiment of the present application, and specifically:

the server may include components such as a processor 401 of one or more processing cores, memory 402 of one or more computer-readable storage media, Radio Frequency (RF) circuitry 403, a power supply 404, an input unit 405, and a display unit 406. Those skilled in the art will appreciate that the server architecture shown in FIG. 12 is not meant to be limiting, and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components. Wherein:

the processor 401 is a control center of the server, connects various parts of the entire server using various interfaces and lines, and performs various functions of the server and processes data by running or executing software programs and/or modules stored in the memory 402 and calling data stored in the memory 402, thereby performing overall monitoring of the server. Optionally, processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor, which mainly handles operating systems, user interfaces, application programs, etc., and a modem processor, which mainly handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 401.

The memory 402 may be used to store software programs and modules, and the processor 401 executes various functional applications and data processing by operating the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data created according to the use of the server, and the like. Further, the memory 402 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 access to the memory 402.

The RF circuit 403 may be used for receiving and transmitting signals during information transmission and reception, and in particular, for receiving downlink information of a base station and then processing the received downlink information by the one or more processors 401; in addition, data relating to uplink is transmitted to the base station. In general, the RF circuitry 403 includes, but is not limited to, an antenna, at least one Amplifier, a tuner, one or more oscillators, a Subscriber Identity Module (SIM) card, a transceiver, a coupler, a Low Noise Amplifier (LNA), a duplexer, and the like. In addition, the RF circuitry 403 may also communicate with networks and other devices via wireless communications. The wireless communication may use any communication standard or protocol, including but not limited to Global System for Mobile communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Message Service (SMS), and the like.

The server also includes a power supply 404 (e.g., a battery) for powering the various components, and preferably, the power supply 404 is logically connected to the processor 401 via a power management system, so that functions such as managing charging, discharging, and power consumption are performed via the power management system. The power supply 404 may also include any component of one or more dc or ac power sources, recharging systems, power failure detection circuitry, power converters or inverters, power status indicators, and the like.

The server may further include an input unit 405, and the input unit 405 may be used to receive input numeric or character information and generate a keyboard, mouse, joystick, optical or trackball signal input in relation to user settings and function control. Specifically, in one particular embodiment, input unit 405 may include a touch-sensitive surface as well as other input devices. The touch-sensitive surface, also referred to as a touch display screen or a touch pad, may collect touch operations by a user (e.g., operations by a user on or near the touch-sensitive surface using a finger, a stylus, or any other suitable object or attachment) thereon or nearby, and drive the corresponding connection device according to a predetermined program. Alternatively, the touch sensitive surface may comprise two parts, a touch detection means and a touch controller. The touch detection device detects the touch direction of a user, detects a signal brought by touch operation and transmits the signal to the touch controller; the touch controller receives touch information from the touch sensing device, converts the touch information into touch point coordinates, sends the touch point coordinates to the processor 401, and can receive and execute commands sent by the processor 401. In addition, touch sensitive surfaces may be implemented using various types of resistive, capacitive, infrared, and surface acoustic waves. The input unit 405 may include other input devices in addition to the touch-sensitive surface. In particular, other input devices may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and the like.

The server may also include a display unit 406, and the display unit 406 may be used to display information input by or provided to the user as well as various graphical user interfaces of the server, which may be made up of graphics, text, icons, video, and any combination thereof. The Display unit 406 may include a Display panel, and optionally, the Display panel may be configured in the form of a Liquid Crystal Display (LCD), an Organic Light-Emitting Diode (OLED), or the like. Further, the touch-sensitive surface may overlay the display panel, and when a touch operation is detected on or near the touch-sensitive surface, the touch operation is transmitted to the processor 401 to determine the type of the touch event, and then the processor 401 provides a corresponding visual output on the display panel according to the type of the touch event. Although in FIG. 12 the touch sensitive surface and the display panel are two separate components to implement input and output functions, in some embodiments the touch sensitive surface may be integrated with the display panel to implement input and output functions.

Although not shown, the server may further include a camera, a bluetooth module, etc., which will not be described herein. Specifically, in this embodiment, the processor 401 in the server loads the executable file corresponding to the process of one or more application programs into the memory 402 according to the following instructions, and the processor 401 runs the application program stored in the memory 402, thereby implementing various functions as follows:

The server can realize the effective effect that any one of the game result prediction devices provided in the embodiments of the present application can realize, which is detailed in the foregoing embodiments and will not be described herein again.

Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable storage medium, and the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.

The foregoing describes in detail a method, an apparatus, and a storage medium for predicting a game result provided in an embodiment of the present application, and a specific example is applied to explain the principle and the implementation of the present application, and the description of the foregoing embodiment is only used to help understand the method and the core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.

Claims

1. A method for predicting game outcome, comprising:

determining a plurality of game feature data sets of each of at least two competitors from game image frames at a predicted time point, wherein each game feature data set comprises at least one game feature data, and different game feature data sets belong to different feature types;

determining an independent prediction win ratio value of each competitor according to all the game feature data groups of the same feature type, wherein the different feature types of the same competitor correspond to one independent prediction win ratio value respectively;

determining a transient weight value corresponding to each feature type according to the prediction time point;

generating a game result according to all the independent prediction win value and the instant weight value, wherein the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type.

2. A method for predicting game outcome according to claim 1, wherein the determining an independent predicted win value for each of the competitors based on all of the game feature data sets of the same feature type comprises:

determining a win rate prediction submodel corresponding to each feature type from the trained win rate prediction models;

processing the game feature data group corresponding to the feature type by using the win rate prediction submodel to obtain an independent prediction win rate value of each competitor corresponding to the feature type;

the determining, according to the prediction time point, an instantaneous weight value corresponding to each of the feature types specifically includes:

and determining the instantaneous weight value corresponding to each feature type by using the trained time weight submodel in the win rate prediction model and the prediction time point.

3. A method for predicting game outcome as claimed in claim 2, further comprising, before said determining a win rate prediction submodel corresponding to each of said feature types from the trained win rate prediction models:

acquiring a plurality of finished game videos, wherein each game video comprises at least two historical competitors, and acquiring a final win rate value of each historical competitor in each game video;

extracting game sample image frames of a plurality of selected time points from each game video;

determining a plurality of game sample feature data sets for each of said historical competitors from each of said game sample image frames, each of said game sample feature data sets comprising at least one game sample feature data, different ones of said game sample feature data sets belonging to different ones of said feature types;

and training a preset winning rate prediction model according to the selected time point, the feature type, the final winning rate value and the game sample feature data set to obtain a trained winning rate prediction model.

4. A method for predicting game outcomes as claimed in claim 3 wherein the predetermined odds prediction model comprises a plurality of odds prediction submodels and a time weighting submodel, and wherein each of the different feature types corresponds to one of the odds prediction submodels, and wherein training the predetermined odds prediction model based on the selected time point, the feature type, the final odds value, and the game sample feature data set comprises:

training the corresponding winning rate prediction submodel according to all the game sample feature data groups of the same feature type and the final winning rate value;

and training the time weight submodel according to the game sample feature data groups of all the feature types of all the selected time points and the final win probability value.

5. The method of claim 4, wherein the feature types include an economic feature type, a killing feature type and a defense tower feature type, the plurality of win rate predictor models include a first logistic regression model, a second logistic regression model and a third logistic regression model, and the training of the win rate predictor models according to all the game sample feature data sets of the same feature type and the final win rate value specifically includes:

training the first logistic regression model according to all the game sample feature data sets of the economic feature types and the final win probability value;

training the second logistic regression model according to all the game sample feature data sets of the killing feature types and the final win probability value;

training the third logistic regression model according to all the game sample feature data sets of the defense tower feature types and the final win probability value.

6. The method of claim 4, wherein the feature types include a wilderness resource feature type, a soldier line feature type and a formation feature type, the plurality of win rate predictor models include a first neural network model, a second neural network model and a third neural network model, and the training of the win rate predictor model according to the feature data sets of all the game samples of the same feature type and the final win rate value specifically includes:

training the first neural network model according to all the game sample feature data groups of the wild area resource feature types and the final win probability value;

training the second neural network model according to all the game sample feature data groups of the solidus feature types and the final win probability value;

and training the third neural network model according to all the game sample feature data groups of the formation feature types and the final win probability value.

7. A method of predicting game outcomes as claimed in claim 1, further comprising, after said generating game outcomes based on all of said independently predicted win values and said instantaneous weight values:

generating a corresponding comment text according to the prediction time point, the comprehensive prediction success rate value and the characteristic contribution value;

and providing the comment text to a user in a preset output mode.

8. An apparatus for predicting a game result, comprising:

a first determining module, configured to determine, from a game image frame at a predicted time point, a plurality of game feature data sets of each of at least two competitors, each of the game feature data sets including at least one game feature data, different game feature data sets belonging to different feature types;

a second determining module, configured to determine, according to all the game feature data sets of the same feature type, an independent predicted win ratio value of each of the competitors, where different feature types of the same competitor correspond to one of the independent predicted win ratio values;

a third determining module, configured to determine, according to the predicted time point, an instantaneous weight value corresponding to each feature type;

and the generating module is used for generating a game result according to all the independent prediction win value and the instant weight value, wherein the game result comprises a comprehensive prediction win value of each competitor and a characteristic contribution value corresponding to each characteristic type.

9. A game result prediction apparatus according to claim 8, wherein the second determination module specifically includes:

the determining unit is used for determining a winning rate prediction submodel corresponding to each feature type from the trained winning rate prediction models;

the processing unit is used for processing the game feature data group corresponding to the feature type by utilizing the win rate forecasting submodel to obtain an independent forecasting win rate value of each competitor corresponding to the feature type;

the third determining unit is specifically configured to:

10. A computer-readable storage medium having stored thereon a plurality of instructions adapted to be loaded by a processor to perform a method of predicting a game outcome as claimed in any one of claims 1 to 7.