CN113058253B - Match prediction method and device for modeling cooperative competition effect - Google Patents
Match prediction method and device for modeling cooperative competition effect Download PDFInfo
- Publication number
- CN113058253B CN113058253B CN202110319873.9A CN202110319873A CN113058253B CN 113058253 B CN113058253 B CN 113058253B CN 202110319873 A CN202110319873 A CN 202110319873A CN 113058253 B CN113058253 B CN 113058253B
- Authority
- CN
- China
- Prior art keywords
- competition
- layer
- team
- cooperation
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
- A63B71/00—Games or sports accessories not covered in groups A63B1/00 - A63B69/00
- A63B71/06—Indicating or scoring devices for games or players, or for other sports activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Physical Education & Sports Medicine (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a competition prediction method and a competition prediction device for modeling a competition effect, and provides a competition model which can learn the competition effect and is used for predicting a competition result. Specifically, after the roles of both teams in the competition to be predicted are obtained, the roles of the individual members can be projected to a hidden space through the competition cooperation model, and further, the respective ability values of the two teams can be accurately predicted through the cooperation effect between modeled teammates and the competition effect between opponents, so that the win and loss results are determined. The invention can accurately predict the result of the match, and the experimental results of a plurality of electronic competitions show that the invention is superior to other methods. Moreover, the invention can be easily generalized to other tasks, such as team formation, formation configuration optimization, MOBA game balance detection.
Description
Technical Field
The invention relates to the technical field of machine learning, artificial intelligence and match data mining, in particular to a match prediction method and device for modeling cooperative competition effect.
Background
Team confrontation is a common form of competition, and common team confrontation games include basketball, football, and the MOBA games DOTA, LOL. Predicting the outcome of a team competition is a challenging task.
Because people have social attributes, members in a game inevitably interact with other members, affecting the outcome of the game. Existing work has focused primarily on the individual capabilities of the members of the modeling team, or on interactions within the modeling team. However, there are a variety of complex interactions in team games, including both team internal interactions (i.e., cooperative effects) and team-to-team interactions (i.e., competitive effects). Meanwhile, members with different importance can be concerned to different degrees in team games, and the influence factor is considered in the fresh work.
Disclosure of Invention
In view of the above, in order to solve the above problems, the present invention provides a competition prediction method and apparatus for modeling a cooperative competition effect, and the technical scheme is as follows:
a game prediction method that models cooperative competition effects, the method comprising:
obtaining the roles of both teams in a competition to be predicted;
calling a competition cooperation effect model obtained by pre-training, wherein the competition cooperation effect model can project the roles of the teams to a hidden space and model the cooperation effect inside the teams and the competition effect among the teams;
aiming at each team in the competition to be predicted, inputting the role of the team into the competition and cooperation effect model to obtain the ability value of the team output by the competition and cooperation model, wherein the ability value consists of the individual ability value of a team member, the overall cooperation ability value of the team and the overall competition ability value of the team;
and determining the win-lose result of the competition to be predicted according to the capacity values of the teams of the two parties in the competition to be predicted.
Preferably, the training process of the competition cooperation model includes:
constructing a basic model, wherein the basic model comprises a feature mapping layer for role projection, a cooperation module for modeling cooperation effect and a competition module for modeling competition effect; wherein the content of the first and second substances,
the characteristic mapping layer is used for extracting role characteristics of team members;
the cooperation module comprises a first feature layer, a first interaction layer and a first gathering layer, wherein the first feature layer is used for extracting team member cooperation feature vectors based on role features, the first interaction layer is used for carrying out pairwise interaction on the team member cooperation feature vectors to obtain cooperation values among the team members, and the first gathering layer is used for gathering the cooperation values among the team members;
the competition module comprises a second characteristic layer, a second interaction layer and a second gathering layer, wherein the second characteristic layer is used for extracting the advantage characteristic vector of a team member of a target party and the weakness characteristic vector of a team member of an opposite party based on role characteristics, the second interaction layer is used for carrying out pairwise interaction on the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party to obtain a competition value of the team member of the target party, and the second gathering layer is used for gathering the competition value of the team member of the target party;
taking the roles of the teams of the two parties in the historical competition as sample data, taking the predicted outcome of the basic model on the sample data approaching to the actual outcome of the historical competition as a target, and adjusting the parameters of each layer in the basic model, wherein the predicted outcome is determined based on the capability value output by the teams of the two parties in the historical competition by the basic model;
and taking the basic model with the adjusted parameters as the competition cooperation model.
Preferably, the first interaction layer comprises a first neural network layer and a first attention mechanism layer;
the first neural network layer is used for interacting every two cooperation characteristic vectors of team members to obtain cooperation values among the team members;
the first attention mechanism layer is used for outputting cooperative attention values among team members;
accordingly, the first aggregation layer is used for aggregating collaboration values among team members based on the collaborative attention value.
Preferably, the second interaction layer comprises a second neural network layer and a second attention mechanism layer;
the second neural network layer is used for carrying out pairwise interaction on the dominant feature vector of the target party team member and the weak feature vector of the opposite party team member to obtain a competition value of the target party team member;
the second attention mechanism layer is used for outputting a competitive attention value between the target party team member and the opposite party team member;
correspondingly, the second aggregation layer is used for aggregating the competition values of the team members of the target party based on the competition attention values.
Preferably, the method for acquiring the sample data includes:
capturing match record data of an online multi-player competition through a web crawler, wherein the match record data comprises roles and actual win and lose results of two teams in the online multi-player competition;
and preprocessing the game record data, wherein the preprocessing comprises game filtering and random sampling of samples.
A game prediction apparatus that models a cooperative competition effect, the apparatus comprising:
the model training unit is used for training to obtain a competition cooperation effect model, and the competition cooperation effect model can project the roles of the teams to a hidden space and model the cooperation effect inside the teams and the competition effect among the teams;
the competition prediction unit is used for obtaining the roles of the two teams in the competition to be predicted; calling the competition cooperation effect model; aiming at each team in the competition to be predicted, inputting the role of the team into the competition and cooperation effect model to obtain the ability value of the team output by the competition and cooperation model, wherein the ability value consists of the individual ability value of a team member, the overall cooperation ability value of the team and the overall competition ability value of the team; and determining the win-lose result of the competition to be predicted according to the capacity values of the teams of the two parties in the competition to be predicted.
Preferably, the model training unit is specifically configured to:
constructing a basic model, wherein the basic model comprises a feature mapping layer for role projection, a cooperation module for modeling cooperation effect and a competition module for modeling competition effect; the characteristic mapping layer is used for extracting role characteristics of team members; the cooperation module comprises a first feature layer, a first interaction layer and a first gathering layer, wherein the first feature layer is used for extracting team member cooperation feature vectors based on role features, the first interaction layer is used for carrying out pairwise interaction on the team member cooperation feature vectors to obtain cooperation values among the team members, and the first gathering layer is used for gathering the cooperation values among the team members; the competition module comprises a second characteristic layer, a second interaction layer and a second gathering layer, wherein the second characteristic layer is used for extracting the advantage characteristic vector of a team member of a target party and the weakness characteristic vector of a team member of an opposite party based on role characteristics, the second interaction layer is used for carrying out pairwise interaction on the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party to obtain a competition value of the team member of the target party, and the second gathering layer is used for gathering the competition value of the team member of the target party; taking the roles of the teams of the two parties in the historical competition as sample data, taking the predicted outcome of the basic model on the sample data approaching to the actual outcome of the historical competition as a target, and adjusting the parameters of each layer in the basic model, wherein the predicted outcome is determined based on the capability value output by the teams of the two parties in the historical competition by the basic model; and taking the basic model with the adjusted parameters as the competition cooperation model.
Preferably, the first interaction layer comprises a first neural network layer and a first attention mechanism layer;
the first neural network layer is used for interacting every two cooperation characteristic vectors of team members to obtain cooperation values among the team members;
the first attention mechanism layer is used for outputting cooperative attention values among team members;
accordingly, the first aggregation layer is used for aggregating collaboration values among team members based on the collaborative attention value.
Preferably, the second interaction layer comprises a second neural network layer and a second attention mechanism layer;
the second neural network layer is used for carrying out pairwise interaction on the dominant feature vector of the target party team member and the weak feature vector of the opposite party team member to obtain a competition value of the target party team member;
the second attention mechanism layer is used for outputting a competitive attention value between the target party team member and the opposite party team member;
correspondingly, the second aggregation layer is used for aggregating the competition values of the team members of the target party based on the competition attention values.
Preferably, the mode for acquiring the sample data by the model training unit includes:
capturing match record data of an online multi-player competition through a web crawler, wherein the match record data comprises roles and actual win and lose results of two teams in the online multi-player competition; and preprocessing the game record data, wherein the preprocessing comprises game filtering and random sampling of samples.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a competition prediction method and a competition prediction device for modeling a competition effect, and provides a competition model which can learn the competition effect and is used for predicting a competition result. Specifically, after the roles of both teams in the competition to be predicted are obtained, the roles of the individual members can be projected to a hidden space through the competition cooperation model, and further, the respective ability values of the two teams can be accurately predicted through the cooperation effect between modeled teammates and the competition effect between opponents, so that the win and loss results are determined. The invention can accurately predict the result of the match, and the experimental results of a plurality of electronic competitions show that the invention is superior to other methods. Moreover, the invention can be easily generalized to other tasks, such as team formation, formation configuration optimization, MOBA game balance detection.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a schematic diagram of a team competition provided by the present invention;
FIG. 2 is a flow chart of a method of game prediction for modeling cooperative competition effects provided by the present invention;
FIG. 3 is a schematic diagram of a collaboration module provided by the present invention;
FIG. 4 is a schematic diagram of a competition module provided in the present invention;
FIG. 5 is a schematic diagram of a competitive cooperation effect model provided by the present invention;
fig. 6 is a schematic structural diagram of a competition prediction device for modeling a cooperative competition effect according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, 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 invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
For ease of understanding of the present solution, the following first describes the cooperative and competitive effects:
cooperative and competitive effects are widely present in group games, such as: teammates will help each other and will compete against each other. Fig. 1 is a schematic diagram of a team competition provided by the present invention.
There are different cooperative and competitive effects between different roles. Taking the hero alliance of the well-known MOBA game as an example, hero in the game can be roughly classified into ADC (output core), auxiliary, meat shield, and guest. In the same team, the advantages of the ADC type hero and the auxiliary type hero can be complemented, so that the competition rate is improved; however, the match between the ADC hero and the ADC hero is poor, which may reduce the game success rate. When two teams compete, the guest hero controls the ADC hero, which means that the guest hero competes with the ADC hero with greater advantage that can increase the success rate of the team where the guest hero is located.
Compared with the traditional method which does not consider interaction or only considers part of interaction, the competition prediction scheme for modeling the cooperative competition effect can model the interaction influence (cooperation and competition relationship) among individuals in the competition in detail. The method has a certain practical application value, not only can be used for predicting the win or loss of the match, but also can help a coach to select roles and optimize team formation configuration. Bringing certain potential economic benefits.
Referring to the method flowchart shown in fig. 2, an embodiment of the present invention provides a competition prediction method for modeling a cooperative competition effect, including the following steps:
and S10, acquiring the roles of the two teams in the game to be predicted.
And S20, calling a competition cooperation effect model obtained by pre-training, wherein the competition cooperation effect model can project the role of the team to a hidden space and model the cooperation effect inside the team and the competition effect among the teams.
In the embodiment of the present invention, the competitive Cooperation effect model may be a Neural access control model (Neural attribute cooperative competition model), which is capable of learning the competitive Cooperation effect and used for predicting the competition result.
Specifically, the NeuralAC model projects the roles of individual members into multiple hidden spaces, using neural networks as interaction functions to model the cooperative effect between teammates and the competitive effect between opponents.
In the concrete implementation process, the training process of the competition cooperation model comprises the following steps:
constructing a basic model, wherein the basic model comprises a feature mapping layer for role projection, a cooperation module for modeling cooperation effect and a competition module for modeling competition effect; wherein the content of the first and second substances,
the characteristic mapping layer is used for extracting role characteristics of the team members; the cooperation module comprises a first characteristic layer, a first interaction layer and a first gathering layer, wherein the first characteristic layer is used for extracting team member cooperation characteristic vectors based on role characteristics, the first interaction layer is used for carrying out pairwise interaction on the team member cooperation characteristic vectors to obtain cooperation values among the team members, and the first gathering layer is used for gathering the cooperation values among the team members; the competition module comprises a second characteristic layer, a second interaction layer and a second gathering layer, the second characteristic layer is used for extracting the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party based on the role characteristics, the second interaction layer is used for carrying out pairwise interaction on the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party to obtain a competition value of the team member of the target party, and the second gathering layer is used for gathering the competition value of the team member of the target party;
the method comprises the steps that the roles of two teams in the historical competition are used as sample data, the goal that the predicted win-lose result of the sample data by a basic model approaches to the actual win-lose result of the historical competition is taken as the target, parameters of all layers in the basic model are adjusted, and the predicted win-lose result is determined based on the capability value output by the two teams in the historical competition by the basic model;
and taking the basic model with the parameter adjustment finished as a competition cooperation model.
In the embodiment of the invention, for the feature mapping layer, the role features of individual members, namely team members, can be extracted by projecting the roles to a plurality of hidden spaces.
For the cooperation module, because different roles have different cooperation characteristics, the role characteristics of each team member are projected to the cooperation vector space, the team members interact with each other pairwise, then the complex cooperation relationship between the team members is captured by using a neural network, the cooperation values of the two team members are obtained, and finally all the cooperation values are collected to obtain the cooperation value of the team.
For the competition module, because different roles have different advantages and weaknesses, the role characteristics of the team members of both sides of the competition are projected to the weak vector space and the dominant vector space, so that the competitors interact with each other pairwise, then the neural network is used for capturing the complex competition relationship between the competitors to obtain the competition values of the team members of both sides of the competition, and finally all the competition values are collected to obtain the competition value of the team.
It should be noted that, when calculating the competition value, for any team of the teams of the two parties of the competition, i.e. the target party team, when calculating the competition value, the role features of the members of the target party team are projected to the dominant vector space, and the role features of the members of the opposite party team are projected to the disadvantaged vector space.
In addition, key characters may have a greater impact on the outcome of the game when teams cooperate. Therefore, they are more concerned by teammates in the game. While the two teams compete, those key figures tend to have a greater impact on the outcome of the game. Therefore, they are more concerned by teammates and opponents in the game. Therefore, the invention further models the attention distribution in the cooperation and competition process, thereby enhancing the prediction accuracy and interpretability of the model.
In order for the competitive cooperation effect model to learn the weighted competitive cooperation effect, the invention uses two attention mechanisms to capture the team content and the attention distribution among the teams, which can simultaneously improve the accuracy and the interpretability of the competition prediction. Specifically, the method comprises the following steps:
the first interaction layer comprises a first neural network layer and a first attention mechanism layer; the first neural network layer is used for carrying out pairwise interaction on the cooperation characteristic vectors of the team members to obtain cooperation values among the team members; the first attention mechanism layer is used for outputting a cooperative attention value among the team members; accordingly, the first aggregation layer is configured to aggregate collaboration values among team members based on the collaborative attention values.
The second interaction layer comprises a second neural network layer and a second attention mechanism layer; the second neural network layer is used for carrying out pairwise interaction on the dominant feature vector of the target party team member and the weak feature vector of the opposite party team member to obtain a competition value of the target party team member; the second attention mechanism layer is used for outputting a competitive attention value between the target party team member and the opposite party team member; accordingly, the second aggregation layer is configured to aggregate the competition values of the team members of the target party based on the competition attention values.
It should be noted that, when summarizing, the first summarization layer takes the superposition of the element-wise product of the cooperation value among the team members and the cooperation attention value among the team members as the overall cooperation capability value of the team. The second aggregation layer, when aggregating, takes the superposition of the competition value of the team member of the target party and the element-wise product of the competition attention value as the total competition capability value of the team of the target party.
See FIG. 3 for a schematic diagram of the collaboration module. The cooperative Embedding layer is a first characteristic layer, the input of the first characteristic layer is role characteristics of team members, cooperative characteristic vectors of the team members can be extracted through projection to a cooperative vector space, then the first interaction layer interacts the cooperative characteristic vectors of every two team members to obtain Cooperation values among the team members and Cooperation attention values among the team members, and finally the first collecting layer collects the Cooperation values among the team members based on the Cooperation attention values to obtain a total Cooperation capability value.
See fig. 4 for a schematic diagram of the competition module. The Competition Embedding layer is a second characteristic layer, the input of the second characteristic layer is the role characteristics of team members of both parties of a target Competition, the advantage characteristic vector of the team member of the target party and the disadvantage characteristic vector of the team member of the opposite party are obtained by projecting the role characteristics of the team member of the target party to the advantage vector space and projecting the role characteristics of the team member of the opposite party to the disadvantage vector space, the Competition value of the team member of the target party and the Competition attention value of the team member of the opposite party are obtained by pairwise interaction of the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party through the second interaction layer, and finally the second collection layer collects the Competition ability value of the team member of the target party based on the Competition attention value to the team member of the target party to obtain the total Competition ability value of the team member of the target party. The process is a calculation process of the total competitive power value of one party of two parties of the competition, and the same is true of a calculation process of the team of the other party as a target party.
In the above description, the basic model of the competitive cooperation model is described as follows:
the invention can predict the outcome of the competition without depending on the competition details, firstly defines and formalizes the problems:
the competition result prediction is to predict the outcome of the teams of both parties according to the role IDs of the team members of both parties of the competition. Suppose there are n members {1, 2.., n }, M observable races. Each match involvesTwo team competition, the two teams being denoted TAAnd TBThe members of each team are a subset of {1, 2.., n }. The result of this M-field match is denoted as { y1，y2，...，yM}。
In the present invention, we only focus on the case where the result of the match is a lose or win, and we assume that there is no tie. If T is in the mth gameABeat TBThen ym1, otherwise ym0. Given a field TA、TBIn the event of an unobserved match, our goal is to predict the outcome of the match
Sample data is first collected before training the base model. The invention can use the match record data of online multi-player competition as an input data set, and the match record data needs to contain the role IDs and the actual win and loss results of both parties. Such data is for example open source data set (DOTA2, LoL), etc. Specifically, the input data set may be obtained by crawling an API provided by a game company through a web crawler. Because of the need for modeling the cooperative competition effect, the large data volume can improve the generalization performance of the model and bring about more accurate prediction.
Furthermore, to ensure the training effect of the model, the input data set may be further preprocessed. The pretreatment mainly comprises the following steps:
1) and (3) filtering the game: a time threshold for the game may be set to filter games of too short duration to ensure the effectiveness of each game.
2) Randomly sampling samples: in the input data set, random sampling is respectively carried out, and the collected subsets are selected to train the basic model.
Further, when the basic model is trained, sample data is input into the basic model, and parameters of the model are optimized by adopting cross entropy loss, so that the predicted win-lose result of the basic model is continuously close to the actual win-lose result. The formula for cross entropy loss is expressed as follows:
where M represents the number of training samples.
As a prediction of the model, i.e. TABeat TBThe probability of (c). y is the true outcome of the game. The training goal of the model is to minimize cross-entropy loss
S30, aiming at each team in the competition to be predicted, inputting the role of the team into the competition and cooperation effect model to obtain the ability value of the team output by the competition and cooperation model, wherein the ability value is composed of the individual ability value of the team member, the overall cooperation ability value of the team and the overall competition ability value of the team.
In the embodiment of the invention, no matter the historical competition or the competition to be predicted, the process of calculating the competition result based on the capacity values of the teams of the two competition parties is as follows:
see figure 5 for a schematic diagram of a competitive cooperative effect model. In team games, a more powerful team is more likely to win. Therefore, we assume that each team has a value representing the team's overall ability. Then TABeat TBIs defined as:
wherein S isAIs TACapacity value of SBIs TBThe capacity value of (c). When S isAMuch greater than SBWhen, TBThere is substantially no chance of winning; when S isAAnd SBWhen equal, both sides win with a 50% probability.
As we said before, there are a number of complex interactions in team confrontation, including teammatesCollaboration between and competition between opponents. Thus, team TACan be expressed in the form (T)BThe capacity value calculation method is the same as that of the following steps:
wherein, wiIs a parameter of a competitive cooperation effect model and represents the individual ability of team members, and the individual ability is independent and is not influenced by others. Fcoop(TA) Representative team TATotal cooperative ability value of (1), Fcomp(TA，TB) Represents TAConfrontation of TBThe overall competitive power value of time.
In addition, the expression formula of the cooperation module is as follows:
wherein v isiAnd vjRepresenting the cooperative feature vector, and is a k-dimensional model parameter vector; an element-wise multiplication; f. of1Is a multilayer perceptron for modeling nonlinear cooperative relationships between teammates. f. of1(vi⊙vj) Is a cooperation value representing the cooperative effect between i and j. The more tacit i and j cooperate, the higher the cooperation value of i and j, and the higher the ability value of the team to which i and j belong.
Is a cooperative attention value, the calculation method of the cooperative attention value is as follows:
the expression formula of the competition module is as follows:
wherein p isiIs the dominant feature vector, cjIs a weak feature vector and is a k-dimensional model parameter vector; f. of2Is a multilayer perceptron used for capturing complex competitive relations between opponents. f. of2(pi⊙cj) The output of (c) is a competition value representing the competing effect of i on j. If i has a greater advantage in the face of j, the higher the competition value of i against j, the higher the ability value of the team to which i belongs.
Is a competitive power value, the calculation method of the competitive power value is as follows:
in our implementation, f1And f2Have the same neural network structure. Here we only explain f2The network structure of (1). Let us let the input p of the neural networki⊙cjFor x, the whole process can be expressed as:
z1＝σ(W1x+b1)，
z2＝σ(W2z1+b2)，
……
zL＝σ(WLzL-1+bL)，
Oij＝ReLU(WozL+bo)
wherein, OijIs f2Is output of (2), representsThe competing effect of i on j; a series of W and b are model parameters of the neural network, and σ is the ReLU activation function.
Here we conclude that S takes into account the individual abilities, the cooperative effect within teams, and the competitive effect between teams simultaneouslyACan be expressed as:
and S40, determining the win and lose result of the competition to be predicted according to the ability values of the two teams in the competition to be predicted.
In the embodiment of the invention, the win and loss results of each party team in the competition to be predicted can be calculated according to the formula (2). And will not be described in detail herein.
After the competitive cooperation effect model is trained, the following effects can be realized:
1) after the formation of the teams of the two parties is given, the model can predict the result of the game which does not occur more accurately.
2) Given the formation of the team of my party, the model can provide a pairwise cooperation value between the members of the team of my party, and if the formation of the team of the opposite party is known, the model can also provide a pairwise competition value between the members of the team of my party and the members of the team of the opposite party. This clearly provides a reference for the coach to select a player, or the electronic sports coach to select a hero.
3) The model is able to identify key people in the team. This can also provide a reference for the coach.
According to the scheme, the result of the multi-person competition is predicted by modeling the cooperative effect with the attention distribution and the competitive effect with the attention distribution. Compared with the prior art, the accuracy and the interpretability of the competition victory or defeat prediction are greatly improved.
Based on the competition prediction method for modeling the cooperative competition effect provided by the above embodiment, an embodiment of the present invention provides an apparatus for executing the competition prediction method for modeling the cooperative competition effect, and a schematic structural diagram of the apparatus is shown in fig. 6, and the apparatus includes:
the model training unit 10 is used for training to obtain a competition cooperation effect model, and the competition cooperation effect model can project the roles of the teams to a hidden space and model the cooperation effect inside the teams and the competition effect among the teams;
a competition prediction unit 20, configured to obtain roles of two teams in a competition to be predicted; a competition cooperation effect model is called; inputting the role of each team in the competition to be predicted into a competition cooperation effect model to obtain the ability value of the team output by the competition cooperation model, wherein the ability value consists of the individual ability value of the team member, the overall cooperation ability value of the team and the overall competition ability value of the team; and determining the win-lose result of the competition to be predicted according to the capability values of the teams of the two parties in the competition to be predicted.
The model training unit 10 is specifically configured to:
constructing a basic model, wherein the basic model comprises a feature mapping layer for role projection, a cooperation module for modeling cooperation effect and a competition module for modeling competition effect; the system comprises a characteristic mapping layer, a characteristic mapping layer and a characteristic mapping layer, wherein the characteristic mapping layer is used for extracting role characteristics of team members; the cooperation module comprises a first characteristic layer, a first interaction layer and a first gathering layer, wherein the first characteristic layer is used for extracting team member cooperation characteristic vectors based on role characteristics, the first interaction layer is used for carrying out pairwise interaction on the team member cooperation characteristic vectors to obtain cooperation values among the team members, and the first gathering layer is used for gathering the cooperation values among the team members; the competition module comprises a second characteristic layer, a second interaction layer and a second gathering layer, the second characteristic layer is used for extracting the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party based on the role characteristics, the second interaction layer is used for carrying out pairwise interaction on the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party to obtain a competition value of the team member of the target party, and the second gathering layer is used for gathering the competition value of the team member of the target party; the method comprises the steps that the roles of two teams in the historical competition are used as sample data, the goal that the predicted win-lose result of the sample data by a basic model approaches to the actual win-lose result of the historical competition is taken as the target, parameters of all layers in the basic model are adjusted, and the predicted win-lose result is determined based on the capability value output by the two teams in the historical competition by the basic model; and taking the basic model with the parameter adjustment finished as a competition cooperation model.
Wherein the first interaction layer comprises a first neural network layer and a first attention mechanism layer;
the first neural network layer is used for carrying out pairwise interaction on the cooperation characteristic vectors of the team members to obtain cooperation values among the team members;
the first attention mechanism layer is used for outputting a cooperative attention value among the team members;
accordingly, the first aggregation layer is configured to aggregate collaboration values among team members based on the collaborative attention values.
Wherein the second interaction layer comprises a second neural network layer and a second attention mechanism layer;
the second neural network layer is used for carrying out pairwise interaction on the dominant feature vector of the target party team member and the weak feature vector of the opposite party team member to obtain a competition value of the target party team member;
the second attention mechanism layer is used for outputting a competitive attention value between the target party team member and the opposite party team member;
accordingly, the second aggregation layer is configured to aggregate the competition values of the team members of the target party based on the competition attention values.
The mode of obtaining sample data by the model training unit 10 includes:
capturing match record data of the online multi-player confrontation match through a web crawler, wherein the match record data comprises roles and actual win and lose results of two party teams in the online multi-player confrontation match; preprocessing the match record data, including match filtering and random sampling of samples.
The invention can accurately predict the result of the match, and the experimental results of a plurality of electronic competitions show that the invention is superior to other methods. Moreover, the invention can be easily generalized to other tasks, such as team formation, formation configuration optimization, MOBA game balance detection.
The match prediction method and device for modeling the cooperative competition effect provided by the invention are introduced in detail, the principle and the implementation mode of the invention are explained by applying specific examples, and the description of the examples is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, 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 invention.
It should be noted that, in the present specification, the embodiments are all described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments may be referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
It is further noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include or include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (6)
1. A game prediction method for modeling cooperative competition effects, the method comprising:
obtaining the roles of both teams in a competition to be predicted;
calling a competition cooperation effect model obtained by pre-training, wherein the competition cooperation effect model can project the roles of the teams to a hidden space and model the cooperation effect inside the teams and the competition effect among the teams;
aiming at each team in the competition to be predicted, inputting the role of the team into the competition and cooperation effect model to obtain the ability value of the team output by the competition and cooperation model, wherein the ability value consists of the individual ability value of the team member, the overall cooperation ability value of the team and the overall competition ability value of the team;
determining the win-lose result of the competition to be predicted according to the capacity values of the teams of the two parties in the competition to be predicted;
the training process of the competition cooperation model comprises the following steps:
constructing a basic model, wherein the basic model comprises a feature mapping layer for role projection, a cooperation module for modeling cooperation effect and a competition module for modeling competition effect; wherein the content of the first and second substances,
the characteristic mapping layer is used for extracting role characteristics of team members;
the cooperation module comprises a first feature layer, a first interaction layer and a first gathering layer, wherein the first feature layer is used for extracting team member cooperation feature vectors based on role features, the first interaction layer is used for carrying out pairwise interaction on the team member cooperation feature vectors to obtain cooperation values among the team members, and the first gathering layer is used for gathering the cooperation values among the team members;
the competition module comprises a second characteristic layer, a second interaction layer and a second gathering layer, wherein the second characteristic layer is used for extracting the advantage characteristic vector of a team member of a target party and the weakness characteristic vector of a team member of an opposite party based on role characteristics, the second interaction layer is used for carrying out pairwise interaction on the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party to obtain a competition value of the team member of the target party, and the second gathering layer is used for gathering the competition value of the team member of the target party;
taking the roles of the teams of the two parties in the historical competition as sample data, taking the predicted outcome of the basic model on the sample data approaching to the actual outcome of the historical competition as a target, and adjusting the parameters of each layer in the basic model, wherein the predicted outcome is determined based on the capability value output by the teams of the two parties in the historical competition by the basic model;
taking the basic model with the adjusted parameters as the competitive cooperation model;
the first interaction layer comprises a first neural network layer and a first attention mechanism layer;
the first neural network layer is used for interacting every two cooperation characteristic vectors of team members to obtain cooperation values among the team members;
the first attention mechanism layer is used for outputting cooperative attention values among team members;
accordingly, the first aggregation layer is used for aggregating collaboration values among team members based on the collaborative attention value.
2. The method of claim 1, wherein the second interaction layer comprises a second neural network layer and a second attention mechanism layer;
the second neural network layer is used for carrying out pairwise interaction on the dominant feature vector of the target party team member and the weak feature vector of the opposite party team member to obtain a competition value of the target party team member;
the second attention mechanism layer is used for outputting a competitive attention value between the target party team member and the opposite party team member;
correspondingly, the second aggregation layer is used for aggregating the competition values of the team members of the target party based on the competition attention values.
3. The method of claim 1, wherein the sample data obtaining means comprises:
capturing match record data of an online multi-player competition through a web crawler, wherein the match record data comprises roles and actual win and lose results of two teams in the online multi-player competition;
and preprocessing the game record data, wherein the preprocessing comprises game filtering and random sampling of samples.
4. A game prediction apparatus for modeling cooperative competition effects, the apparatus comprising:
the model training unit is used for training to obtain a competition cooperation effect model, and the competition cooperation effect model can project the roles of the teams to a hidden space and model the cooperation effect inside the teams and the competition effect among the teams;
the competition prediction unit is used for obtaining the roles of the two teams in the competition to be predicted; calling the competition cooperation effect model; aiming at each team in the competition to be predicted, inputting the role of the team into the competition and cooperation effect model to obtain the ability value of the team output by the competition and cooperation model, wherein the ability value consists of the individual ability value of the team member, the overall cooperation ability value of the team and the overall competition ability value of the team; determining the win-lose result of the competition to be predicted according to the capacity values of the teams of the two parties in the competition to be predicted;
the model training unit is specifically configured to:
constructing a basic model, wherein the basic model comprises a feature mapping layer for role projection, a cooperation module for modeling cooperation effect and a competition module for modeling competition effect; the characteristic mapping layer is used for extracting role characteristics of team members; the cooperation module comprises a first feature layer, a first interaction layer and a first gathering layer, wherein the first feature layer is used for extracting team member cooperation feature vectors based on role features, the first interaction layer is used for carrying out pairwise interaction on the team member cooperation feature vectors to obtain cooperation values among the team members, and the first gathering layer is used for gathering the cooperation values among the team members; the competition module comprises a second characteristic layer, a second interaction layer and a second gathering layer, wherein the second characteristic layer is used for extracting the advantage characteristic vector of a team member of a target party and the weakness characteristic vector of a team member of an opposite party based on role characteristics, the second interaction layer is used for carrying out pairwise interaction on the advantage characteristic vector of the team member of the target party and the weakness characteristic vector of the team member of the opposite party to obtain a competition value of the team member of the target party, and the second gathering layer is used for gathering the competition value of the team member of the target party; taking the roles of the teams of the two parties in the historical competition as sample data, taking the predicted outcome of the basic model on the sample data approaching to the actual outcome of the historical competition as a target, and adjusting the parameters of each layer in the basic model, wherein the predicted outcome is determined based on the capability value output by the teams of the two parties in the historical competition by the basic model; taking the basic model with the adjusted parameters as the competitive cooperation model;
the first interaction layer comprises a first neural network layer and a first attention mechanism layer;
the first neural network layer is used for interacting every two cooperation characteristic vectors of team members to obtain cooperation values among the team members;
the first attention mechanism layer is used for outputting cooperative attention values among team members;
accordingly, the first aggregation layer is used for aggregating collaboration values among team members based on the collaborative attention value.
5. The apparatus of claim 4, wherein the second interaction layer comprises a second neural network layer and a second attention mechanism layer;
the second neural network layer is used for carrying out pairwise interaction on the dominant feature vector of the target party team member and the weak feature vector of the opposite party team member to obtain a competition value of the target party team member;
the second attention mechanism layer is used for outputting a competitive attention value between the target party team member and the opposite party team member;
correspondingly, the second aggregation layer is used for aggregating the competition values of the team members of the target party based on the competition attention values.
6. The apparatus of claim 4, wherein the manner in which the model training unit obtains the sample data comprises:
capturing match record data of an online multi-player competition through a web crawler, wherein the match record data comprises roles and actual win and lose results of two teams in the online multi-player competition; and preprocessing the game record data, wherein the preprocessing comprises game filtering and random sampling of samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110319873.9A CN113058253B (en) | 2021-03-25 | 2021-03-25 | Match prediction method and device for modeling cooperative competition effect |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110319873.9A CN113058253B (en) | 2021-03-25 | 2021-03-25 | Match prediction method and device for modeling cooperative competition effect |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113058253A CN113058253A (en) | 2021-07-02 |
| CN113058253B true CN113058253B (en) | 2022-03-01 |
Family
ID=76561897
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110319873.9A Active CN113058253B (en) | 2021-03-25 | 2021-03-25 | Match prediction method and device for modeling cooperative competition effect |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113058253B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230277946A1 (en) * | 2021-11-24 | 2023-09-07 | International Business Machines Corporation | Detecting meta-environment changes |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109603159A (en) * | 2018-12-18 | 2019-04-12 | 网易（杭州）网络有限公司 | Match the method, apparatus and system of game player |
| CN110147524A (en) * | 2019-05-10 | 2019-08-20 | 深圳市腾讯计算机系统有限公司 | A kind of result of the match prediction technique, device and equipment based on machine learning |
| CN110478905A (en) * | 2019-08-28 | 2019-11-22 | 中国银行股份有限公司 | Sports object select method and device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5107405B2 (en) * | 2010-08-06 | 2012-12-26 | 株式会社コナミデジタルエンタテインメント | GAME SYSTEM, GAME SYSTEM CONTROL METHOD AND PROGRAM |
| WO2019144146A1 (en) * | 2018-01-21 | 2019-07-25 | Stats Llc | Method and system for interactive, interpretable, and improved match and player performance predictions in team sports |
-
2021
- 2021-03-25 CN CN202110319873.9A patent/CN113058253B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109603159A (en) * | 2018-12-18 | 2019-04-12 | 网易（杭州）网络有限公司 | Match the method, apparatus and system of game player |
| CN110147524A (en) * | 2019-05-10 | 2019-08-20 | 深圳市腾讯计算机系统有限公司 | A kind of result of the match prediction technique, device and equipment based on machine learning |
| CN110478905A (en) * | 2019-08-28 | 2019-11-22 | 中国银行股份有限公司 | Sports object select method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113058253A (en) | 2021-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Valero | Predicting Win-Loss outcomes in MLB regular season games–A comparative study using data mining methods | |
| Ortega et al. | Imitating human playing styles in super mario bros | |
| CN107158708A (en) | Multi-player video game matching optimization | |
| CN113660987A (en) | Generating gaming robots adapted for gaming applications | |
| Tao et al. | Mvan: Multi-view attention networks for real money trading detection in online games | |
| Lee et al. | Learning a Super Mario controller from examples of human play | |
| CN113058253B (en) | Match prediction method and device for modeling cooperative competition effect | |
| Thawonmas et al. | Artificial general intelligence in games: Where play meets design and user experience | |
| Eger et al. | Deck archetype prediction in hearthstone | |
| Rezvani et al. | Gamers' Behaviour and Communication Analysis in Massively Multiplayer Online Games: A Survey | |
| Zhao et al. | Winning tracker: a new model for real-time winning prediction in MOBA games | |
| Barthet et al. | Go-blend behavior and affect | |
| Smith et al. | Mining controller inputs to understand gameplay | |
| Cavadenti et al. | When cyberathletes conceal their game: Clustering confusion matrices to identify avatar aliases | |
| CN115944921A (en) | Game data processing method, device, equipment and medium | |
| King | Predicting a Quarterback's fantasy football point output for daily fantasy sports using statistical models | |
| McEachern | Game theory: A classical introduction, mathematical games, and the tournament | |
| Chesani et al. | Can Deep Networks Learn to Play by the Rules? A Case Study on Nine Men's Morris | |
| Chazan-Pantzalis | Sports Analytics Algorithms for Performance Prediction | |
| Beheshtian-Ardakani et al. | CMPN: Modeling and analysis of soccer teams using Complex Multiplex Passing Network | |
| Lee et al. | A novel ontology for computer Go knowledge management | |
| Wang et al. | [Retracted] Influence of Different Passing Methods of Physical Fitness in Football Using Deep Learning | |
| Cowley et al. | Short Literature Review for a General Player Model Based on Behavlets | |
| Vargas et al. | Active learning and case-based reasoning for the deceptive play in the card game of truco | |
| Dockhorn et al. | Association Rule Mining for Unknown Video Games |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |