Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the application.
The term "and/or" is used herein to describe association of associated objects, and specifically indicates that three relationships may exist, for example, a and/or B may indicate: a exists alone, A and B exist together, and B exists alone.
It should be noted that, the user information (including but not limited to user equipment information, user personal information, etc.) and the data (including but not limited to data for analysis, stored data, presented data, etc.) related to the present application are information and data authorized by the user or fully authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related country and region, and provide corresponding operation entries for the user to select authorization or rejection.
In order to clearly describe the technical solution of the embodiments of the present application, firstly, the terms involved in the present application are explained:
SLG: the Game of the strategy class.
MMO: massive (Massively) Multiplayer Online Game, massive multiplayer online game.
And (3) taking together: means that the data of a plurality of servers are combined to form a new server.
PVE: player versus environment, player engagement environment, is a game engagement mode.
PVP: player VS Player, player against Player.
DNN: deep Neural Networks, deep neural network.
pair-wise: paired.
Dim: dimension.
server: and a server.
Label: and (5) a label.
As shown in fig. 1, the present embodiment provides an electronic apparatus 1 including: at least one processor 11 and a memory 12, one processor being exemplified in fig. 1. The processor 11 and the memory 12 are connected by a bus 10. The memory 12 stores instructions executable by the processor 11, and the instructions are executed by the processor 11, so that the electronic device 1 can execute all or part of the methods in the embodiments described below, so as to effectively perform data enhancement on a small amount of historical fit data, and the effect of automatically predicting multiple fit schemes by using limited historical fit data can be achieved, so that the prediction accuracy of the server fit scheme is improved.
In an embodiment, the electronic device 1 may be a mobile phone, a tablet computer, a notebook computer, a desktop computer, or a large computing system composed of a plurality of computers.
Fig. 2 is a schematic diagram of an application scenario 200 of a server merging result prediction system according to an embodiment of the present application. As shown in fig. 2, the system includes: server 210 and terminal 220, wherein:
the server 210 may be a data platform that provides a server merge result prediction service, such as an electronic game compliance platform. In a practical scenario, an electronic game service platform may have a plurality of servers 210, for example, 1 server 210 in fig. 2.
The terminal 220 may be a computer, a mobile phone, a tablet, or other devices used when the user logs in to the electronic game service platform, or a plurality of terminals 220 may be provided, and 2 terminals 220 are illustrated in fig. 2 as an example.
Information transmission between the terminal 220 and the server 210 may be performed through the internet, so that the terminal 220 may access data on the server 210. The terminal 220 and/or the server 210 may be implemented by the electronic device 1.
The server merging result prediction scheme in the embodiment of the application can be deployed on the server 210, the terminal 220 or the server 210 and the terminal 220. The actual scene may be selected based on actual requirements, which is not limited in this embodiment.
When the server merge result prediction scheme is deployed in whole or in part on the server 210, an interface may be invoked open to the terminal 220 to provide algorithmic support to the terminal 220.
The method provided by the embodiment of the application can be realized by the electronic equipment 1 executing corresponding software codes and by carrying out data interaction with a server. The electronic device 1 may be a local terminal device. When the method is run on a server, the method can be implemented and executed based on a cloud interaction system, wherein the cloud interaction system comprises the server and the client device.
In a possible implementation manner, the method provided by the embodiment of the present application provides a graphical user interface through a terminal device, where the terminal device may be the aforementioned local terminal device or the aforementioned client device in the cloud interaction system.
The server merging result prediction mode of the embodiment of the application can be applied to any field needing server merging.
Taking a game server merging scene as an example, an electronic game refers to all interactive games running by relying on an electronic equipment platform. With the development of the internet, more and more electronic games rely on server packages to run and process game data. Taking online electronic games as an example, the life cycle of a game server generally comprises the stages of opening clothes, rising people, peak people, falling people, aggravating player loss, stopping clothes and the like, and the online number of players of the game server is a key for maintaining the vitality of the game.
To maintain game vitality and player experience, game operations typically choose to perform server consolidation in a steadily decreasing stage before players have lost a lot: two or more existing game servers are selected to be combined to obtain a new game server, so that the combined server is restored to a state capable of operating normally. For racing season games, the combination of game servers is a core mechanism for promoting the forward development of the games, and is a key problem of ecological evolution of the games.
However, it is not easy to choose the optimal game server merging scheme, and in a practical scenario, each game server is a complex of a large number of individuals, groups and organizations connected through a complex relationship network, so that a relatively stable benefit group can preempt limited resources, such as alliance, in the game server. Temporary teams are also available to connect small groups of players for personal participation and completion of tasks, such as teams, buddy groups, etc. The game mechanism is complex and changeable, and the factors influencing the combined service result are numerous. The existing server has less history merging data and very sparse data, the ecology of the game is always changed along with the time, the rules of the existing server are difficult to learn through simple technical schemes such as statistics and similarity calculation according to the existing history merging data, and the effect of the merging scheme cannot be accurately predicted.
In order to solve the problems, the embodiment of the application provides a game server merging effect prediction scheme based on deep learning, which comprises the steps of firstly determining a plurality of merging schemes to be predicted, determining the prediction characteristics of a new server after synthesis through the characteristic data of an old server in each merging scheme, combining the plurality of merging schemes two by two, generating comparison characteristics by the prediction characteristics corresponding to the two merging schemes in each combination, inputting the comparison characteristics of the plurality of combinations into a preset sequencing model to obtain the sequencing result of each merging scheme, wherein the preset sequencing model is obtained by adopting the comparison sample characteristics of the historical merging schemes, so that a small amount of historical merging data is effectively subjected to data enhancement, the effect of the plurality of merging schemes can be automatically predicted by using limited historical merging data, and the prediction accuracy of the server merging schemes is improved. I.e., predicting possible server operating states after server consolidation in advance, thereby facilitating operators to select proper occasions and proper combinations for server consolidation.
Compared with the traditional personal experience based on operators, the scheme is more controllable for the state after the combination, and has the main advantages that:
(1) From game player's angle, ensured the ecological balance nature of recreation better, promoted player experience.
(2) From the angle of developer, reduce ecological operation personnel threshold, promote operating efficiency, reduce the resource input, effectively promote the user and remain and the player is active etc. operation index.
Some embodiments of the present application are described in detail below with reference to the accompanying drawings. In the case where there is no conflict between the embodiments, the following embodiments and features in the embodiments may be combined with each other. In addition, the sequence of steps in the method embodiments described below is only an example and is not strictly limited.
Please refer to fig. 3, which is a method for predicting a server merge result according to an embodiment of the present application, wherein the method can be executed by the electronic device 1 shown in fig. 1 and can be applied to an application scenario of the server merge result prediction shown in fig. 2, so as to effectively enhance data of a small amount of historical merge data, automatically predict effects of a plurality of merge schemes by using limited historical merge data, and improve prediction accuracy of the server merge scheme. In this embodiment, taking the terminal 220 as an executing terminal as an example, the method includes the following steps:
step 301: and obtaining a plurality of to-be-predicted combined service schemes, wherein one combined service scheme comprises a plurality of old servers to be combined.
In this step, the old server refers to a server to be merged that has running data, where the running data may include threads or processes running on the server (such as game processes), user access data, and the like. For example, a birth server in a season game scene, hereinafter referred to as "birth service", may be used as an old server to be combined. The compliance solution refers to a predetermined solution of which old servers are to be combined. In an actual scenario, there are generally multiple old servers that need to be combined, which of the old servers can be determined in advance, for example, 10 old servers that need to be combined, and 3 new servers need to be combined, and then multiple combining schemes can be used, and before implementing these combining schemes, which combining scheme can be predicted to be better by adopting the embodiment of the present application. The multiple compliance schemes to be predicted can be recorded by a worker or automatically read from a cloud or local database.
In the game server merging scene, the core of the merging task is to ensure that the new merged server operates normally, stabilize the activity of players, keep the situation balance and avoid the loss of users. Taking season games as an example, there are two types of server concepts accompanying a player per season (typically around three months): birth and season wear. The birthday is a fixed server attribute of a player, the change of the birthday is not generated after the game is opened (the birthday can be carried out in the later period and is smaller), the player data of a plurality of birthdays need to be combined after each season is finished to form a new season server, the progress of the common game of a plurality of birthday players is started in the new season server, and the prediction key point of the combined service scheme is to measure the running effect of the new season server formed after the combination of the plurality of birthday.
Step 302: and determining the prediction characteristics of the new server formed by the single concurrent service scheme according to the characteristic data corresponding to the plurality of old servers aiming at the single concurrent service scheme.
In this step, there are a plurality of the compliance schemes to be predicted, the purpose of the prediction is to measure the results of the plurality of compliance schemes, and the operation state of the new server formed in each compliance scheme can be used to characterize the merits of the corresponding compliance schemes. Firstly, aiming at each service combination scheme to be predicted, determining the prediction characteristics of a new server formed by a single service combination scheme according to the characteristic data of a plurality of old servers. The characteristic data is used for characterizing the characteristics of the old server in terms of operation data, and can be used for evaluating the operation performance of the old server. The new server is a new server formed by combining data of a plurality of old servers, and the new server has combined data and can independently operate, so that the prediction characteristics of the new server formed by a single service combination scheme can be predicted based on the characteristic data of the plurality of old servers to be combined in the service combination scheme. And for a plurality of to-be-predicted combined service schemes, predicting the prediction characteristics of the corresponding new servers by adopting the mode, so that the prediction characteristics of the corresponding new servers of each combined service scheme can be obtained.
In one embodiment, the step 302 may specifically include: and aiming at the single concurrent service scheme, respectively acquiring the current operation data of a plurality of old servers, and determining the prediction characteristics of the new servers formed by the corresponding single concurrent service scheme according to the current operation data.
In this embodiment, the prediction features are used to characterize the data features of the new server after the combination, so that a quantization standard for evaluating the attribute of the new server needs to be formulated first, and a specific scene can be measured by using different custom indexes. Taking a game server merging scene as an example, in an actual scene, the running data of the game server may include game data, and the prediction features of the new server may include active points and/or situation points presented by the game data, i.e. active points and/or situation points of the new server may be introduced to measure player activity and game situation of the new server after the game server is taken as an example. Specifically, the activity score is used to represent the activity degree of the user in the game, and the factors affecting the activity score may include comprehensive indexes calculated by the online average time length, login times, PVP times, PVE times and the like of the user. For example, the longer the online average duration of a user of a certain server, the greater the number of logins of the user in a certain time, the greater the activity level of the server, and the greater the activity score.
The situation score is used for representing situation distribution information of each player participating in the game, which is presented by the game data in the server, and factors influencing the situation score in the actual scene can comprise: the composite score for each player's potential (e.g., the potential of a different alliance) at a particular point in time (e.g., one month from a new season), e.g., including the number of occupied cells, the number of players, the personal potential value composite, the personal combat composite, etc.
The feature data may include current operation data of the old servers, and for a single compliance solution, current operation data of a plurality of old servers corresponding to the compliance solution may be obtained respectively, where the current operation data may include attribute data and relationship data of the corresponding old servers, and the attribute data is used to characterize player attributes in the corresponding old servers. The relationship data is used to characterize the relationship between different old servers in a single compliance solution. The current operational data includes, but is not limited to, one or more of the number of online people on each old server, the length of the user login, and the winning rate of the game play. And according to the current operation data, estimating the activity score and/or situation score of the new server formed by the combination scheme, and processing other combination schemes to be predicted according to the mode to obtain the activity score and/or situation score of the new server formed by each combination scheme. In this manner, state information for the old server may be collected from the player's perspective such that the predicted features of the new server take into account the player's gaming experience.
In one embodiment, determining the predicted features of the new server formed corresponding to the single compliance program in step 302 according to the current operation data may specifically include: and determining the integral attribute characteristics of the new server formed by the corresponding single compliance solution according to the attribute data of each old server. And obtaining the prediction characteristics of the new server according to the attribute data, the relation data and the overall attribute characteristics of the new server of each old server.
In this embodiment, the current running data includes attribute data and relationship data corresponding to the old server, where the attribute data is used to characterize player attributes within the corresponding old server. The relationship data is used to characterize the relationship between different old servers in a single compliance solution. In an actual game scene, the players and the meetings in the game server have intricate attributes and relations, such as a fight relation and an exchange relation among different birthdays, and the attribute data of the old server and the relation data among different old servers can be determined by modeling the potential features of the servers. Taking the game of the season as an example, each season takes the birth service as an old server, the corresponding season service is a new server after the season service, for example, the server of the first season is obtained by combining a plurality of birth service, and the server of the second season is also obtained by combining a plurality of birth service. For each co-administration scheme, the characteristic data of the birth service may be divided into two parts: birth service attribute features and birth service relationship features. The birth service attribute feature table characterizes the attributes of the player itself within the game, such as the number of players, the number of birth service, the attributes of the player's assessment within the game, and so forth. The birthday attribute features can be divided into overall attribute features and individual birthday attribute features on this basis.
Wherein a single birth service attribute may be determined as follows: the combination regimen is assumed to be: the new season wear is synthesized by 6 births of the births 1, 2, 4, 6, 9 and 10, and the first four births of the players can be selected from the 6 births, and the attribute data of each birthday in the birthday set A is counted by taking the births as a unit under the assumption that the birthday set A of the first four players is selected as the 'birthday 4, 6, 9 and 10'. Each birth service is independently calculated, and for one birth service, the average value of each attribute of a player in the birth service, such as on-line time length, login times, PVP (polyvinyl pyrrolidone) winning rate and the like, can be used as attribute data of the birth service according to the average value of each attribute of a certain birth service in a period of time, for example, the average value of the login times of each player in the past 10 days of the birth service 4 can be used as the login times of each player of the birth service 4 on the assumption that the attribute data of an old server comprises the login times of each player, so that each attribute value is calculated to obtain the attribute characteristics of the single birth service, and the stability and the accuracy of attribute data statistics are improved.
The overall attribute features are some statistical features of the group of birthdays in the set of birthdays a, such as an average, variance, maximum, minimum, etc. of the group of birthday attribute features. For example, the average attribute data of the 6 birthdays in the birthday set A can be obtained, and the average attribute data is used as the integral attribute feature of the new server formed by the service combination scheme.
The relationship data represents relationship attributes among the birth service, including but not limited to fight relationship and exchange relationship among the birth service, contains dependency relationship among the birth service to be combined, and has important influence on combined situation and activity. The relationship data of the birth service in each combined service scheme can be counted, and the corresponding relationship characteristics of the combined service scheme can be determined. The relationship features, attribute features may be represented by feature vectors, such as feature vectors of the relationship features may be at least: vector of three dimensional feature values of fight relationship, alliance relationship and friendly relationship. For each combined service scheme, the feature vector of the corresponding overall attribute feature, the vector of the single birth service attribute feature and the feature vector of the relation feature are spliced to obtain the prediction feature of the new server formed by the combined service scheme, so that the prediction feature fully considers the influence of the data representation of the old server on the combined new server from the angle of a player, and the accuracy of the prediction feature is improved.
Step 303: combining the multiple concurrent taking schemes two by two to obtain at least one scheme pair, wherein the scheme pair comprises a first concurrent taking scheme and a second concurrent taking scheme, the first concurrent taking scheme has a first prediction characteristic, and the second concurrent taking scheme has a second prediction characteristic. And constructing contrast features of the corresponding scheme pairs according to the first prediction features and the second prediction features.
In this step, a plurality of the pair-wise combination schemes to be predicted may be combined, a pair-wise combination feature may be constructed, one or more scheme pairs may be formed, each of the pair-wise combination schemes includes a first combination scheme and a second combination scheme, the first combination scheme has a first prediction feature, the second combination scheme has a second prediction feature, and then a pair-wise comparison feature of the pair-wise combination scheme may be constructed based on the first prediction feature and the second prediction feature, so, assuming that 10 births need to be combined into 2 racing season cases, in order to predict which of the plurality of combination schemes is better, the problem may be converted into a problem of ordering the plurality of combination schemes, by constructing the pair-wise comparison feature, each of the pair-wise combination schemes may obtain an ordering result of the two combination schemes, and the plurality of pairs may obtain an ordering result between the plurality of combination schemes to be predicted.
In one embodiment, step 303 may specifically include: and dividing the first prediction feature and the second prediction feature to obtain the contrast feature of the corresponding scheme pair.
In this embodiment, the prediction features of the new server formed by the combination scheme are determined by attribute data and relationship data of a plurality of old servers, the attribute data and relationship data of the old servers are very many, the corresponding feature vectors can reach 108 dimensions, the data flow is very large, and when the pair bits are generated, the first prediction features and the second prediction features can be divided, so that the data volume of the comparison features is ensured not to be greatly increased, and the calculation efficiency is improved. For example, the feature vector of the first prediction feature of a certain scheme pair is 108 dimensions, the feature vector of the second prediction feature is 108 dimensions, and in an actual scene, the feature vector of the first prediction feature and the feature vector of the second prediction feature can be directly used as the contrast feature of the scheme pair, so that the data volume of the contrast feature at least comprises 216 dimensions, and the data volume is very large. If the feature vector of the first predicted feature is 108 dimensions, the feature vector of the second predicted feature is divided by 108 dimensions, and the obtained contrast feature is also 108 dimensions, so that compared with the contrast feature, the data dimension of the contrast feature is greatly reduced, and the data calculation amount is reduced.
Step 304: and inputting the comparison features of at least one scheme pair into a trained preset sequencing model for identification, and outputting sequencing results of a plurality of combined service schemes.
In this step, the preset ranking model may be a ranking model based on Pair-wise, and the preset ranking model is obtained by training the features of the comparison samples based on a plurality of historical compliance schemes. The comparison sample features are Pair-wise based combined features, the Pair-wise based combined feature method is a Pair-wise based learning method, two data points can be used as a Pair of samples, and the relation between the two data points is learned, so that large-scale historical combined scheme data is not needed to be used as the samples, a small amount of historical combined scheme data can also learn the relative comparison relation of a plurality of combined schemes, data enhancement can be effectively performed on a small amount of historical combined scheme data, and a preset ordering model obtained based on the enhanced historical combined data training is more accurate. The comparison characteristics of the scheme pairs to be predicted are input into the preset sequencing model, a comparison result between two schemes to be predicted in each scheme pair to be predicted can be output, further, a more accurate sequencing result of a plurality of combined service schemes can be output, the sequencing result is used for representing the sequencing of the predicted running states of a plurality of new servers respectively formed by the combined service schemes, for example, the combined service schemes corresponding to the new servers with better predicted running states can be set, and the sequencing result is more forward. Based on the sequencing result output by the preset sequencing model, which combination scheme is better can be determined, so that the effect of automatically predicting the combination scheme of the server is realized by constructing comparison features, and the prediction accuracy of the combination scheme is improved.
Assume that the multiple co-taking schemes to be predicted are: scheme 1, scheme 2 and scheme 3, two-by-two combination can obtain 3 scheme pairs [ scheme 1, scheme 2], [ scheme 1, scheme 3], [ scheme 3, scheme 2], after three scheme pairs are input into the preset ordering model, the comparison relation between two schemes in each scheme pair can be obtained respectively, and it is assumed that in scheme pair [ scheme 1, scheme 2], the result comparison relation is that scheme 1 is better than scheme 2, in scheme pair [ scheme 1, scheme 3] the result comparison relation is that scheme 3 is better than scheme 1, in scheme pair [ scheme 3, scheme 2], the result comparison relation is that scheme 3 is better than scheme 2, in summary, the preset ordering model can output the final ordering result in turn: scheme 3, scheme 2, scheme 1, wherein the more predictive running state the new server corresponding to the compliance scheme is set, the more forward the sequencing result is.
In an embodiment, before step 304, the method may further include: the step of training a predictive ranking model is as follows:
according to a plurality of history service schemes, a plurality of sample server pairs and preset scores of all sample servers are obtained, wherein each sample server pair comprises a first sample server and a second sample server, and each history service scheme comprises a plurality of first old servers combined to form the first sample server and a plurality of second old servers combined to form the second sample server. For a single sample server pair, determining first sample features of a corresponding first sample server according to historical compliance program data of a corresponding plurality of first old servers, and determining second sample features of a second sample server according to historical compliance data of a corresponding plurality of second old servers. And constructing a comparison sample characteristic of the corresponding sample server pair according to the first sample characteristic and the second sample characteristic. And labeling the comparison sample characteristics according to the first preset score of the first sample server and the second preset score of the second sample server, and obtaining labeled comparison sample characteristics of the corresponding sample server pairs. Training a preset sorting network by adopting the labeled sample characteristic comparison pair sample characteristics to generate a prediction sorting model.
In the embodiment, the fact that the data of the history compliance scheme is less is considered, the combination characteristic based on the pair-wise can be constructed, the situation division and active division regression task of the new server is converted into the classification task of wanting to match the running state of the new server after the compliance, and the training data size is greatly improved. The traditional classification task adopts a machine learning method, such as a lifting tree model, a linear regression model and the like, and the embodiment of the application considers the complexity of the contrast characteristic and can adopt a deep learning method to lift the model expression of the complex vector space. The sample server is a server formed after the server merging is completed in the history compliance data. Taking the example of a season game server merging scenario, assume that two completed historical service schemes are:
combination regimen X: the birthday suit 1, the birthday suit 2, the birthday suit 4, the birthday suit 6, the birthday suit 9 and the birthday suit 10 are combined into a racing season suit x. Wherein the racing season suit x belongs to a sample server, and the birth suit 1, the birth suit 2, the birth suit 4, the birth suit 6, the birth suit 9 and the birth suit 10 are a plurality of old servers for synthesizing the racing season suit x.
And (3) a combined taking scheme Y: the birthday clothes 3, the birthday clothes 5, the birthday clothes 7 and the birthday clothes 8 are combined into a racing season clothes y. The racing season clothes y belong to sample servers, and the birth clothes 3, 5, 7 and 8 are a plurality of old servers for synthesizing the racing season clothes y. The season wear x and season wear y may form a sample server pair for constructing the pair-wise based composite feature.
The preset score is a preconfigured score of each sample server, and the preset score is used for characterizing the running data of the sample server, so that the attribute of the sample server can be evaluated, for example, the attribute can be the activity of a player and/or the characteristic of the game situation. In an actual scene, a quantization standard for evaluating the attribute of the new server can be firstly formulated, and the specific scene can be measured by adopting different custom indexes. Taking a game server merging scene as an example, the activity score and/or situation score of a server can be introduced to measure the player activity and the game situation of a new server after the server is matched, for example, the activity score and/or situation score can be used as a preset score. The description of the active and partial points is referred to in the foregoing related description and is not repeated here.
The sample features refer to data features corresponding to a plurality of old servers composing the sample server, and each sample server can obtain the corresponding sample features by counting historical fit data of the corresponding plurality of old servers.
Taking the season-racing game as an example, in the historical clothes-racing data, the birth clothes are old servers, and the season-racing clothes are sample servers after clothes-racing. For each sample server, the corresponding birth service features may be divided into two parts: the birthday sample attribute characteristics and birthday sample relationship characteristics. The birth service sample attribute feature table characterizes the attributes of the player itself within the game, such as the number of players, the number of birth service, the attributes of the player's evaluation within the game, etc. The birthday attribute features can be divided into an overall sample attribute feature and a single birthday sample attribute feature on this basis.
Wherein a single birth service sample attribute feature may be determined as follows: assuming that the sample server x is synthesized by 6 births including a birthlet 1, a birthlet 2, a birthlet 4, a birthlet 6, a birthlet 9 and a birthlet 10, the birthlet with the first player number ranking can be selected from the historical operation data of the 6 birthlets, and the birthlet set B with the first four player numbers ranking is assumed to be 'birthlet 1, birthlet 6, birthlet 9 and birthlet 10', and the attribute data of each birthlet in the birthlet set B is counted by taking the birthlet as a unit. Each birth service is independently calculated, for one birth service, the average value of each attribute of a player in the history running data of the birth service, such as on-line time length, login times, PVP (polyvinyl pyrrolidone) win rate and the like, can be used as attribute data of the birth service according to the average value of each attribute of a certain birth service in a period of time, for example, the average value of the login times of each player of the birth service 4 in the past period of time can be used as the login times of each player of the birth service 4, so that each attribute value is calculated to obtain sample attribute characteristics of a single birth service, and the stability and accuracy of sample attribute data statistics are improved.
The overall sample attribute features are some statistical features of pointers to the group of birthdays in the birthday set B, such as the average, variance, maximum, minimum, etc. of the group of birthday attribute features. For example, the average attribute data of the 6 birthdays in the birthday set B can be obtained, and the average attribute data is used as the integral sample attribute feature of the sample server x.
The sample relationship data represents relationship attributes among the birth service, including but not limited to a fight relationship and an exchange relationship among the birth service, and contains a dependency relationship among the merged birth service, which has important influence on the situation and activity of the sample server x. The relationship data of the birth service of each sample server can be counted to determine the corresponding sample relationship characteristics of the sample server. The method can be expressed in a characteristic vector mode, and is characterized by relation and attribute. For each sample server, the characteristic vector of the corresponding integral sample attribute characteristic, the characteristic vector of the sample attribute characteristic of the single birth service and the characteristic vector of the sample relation characteristic are spliced to obtain the sample characteristic of the sample server, so that the sample characteristic is based on the historical fit data of the old server, the influence of the data representation of the old server on the fit of the new server is fully considered from the player's perspective, and the accuracy of the sample characteristic is improved.
Based on the foregoing principle, for each sample server pair, taking the sample server pair P (season clothes x, season clothes y) as an example, where season clothes x is a first sample server, birth clothes 1, birth clothes 2, birth clothes 4, birth clothes 6, birth clothes 9, and birth clothes 10 are a plurality of first old servers combined to form the season clothes x, and the first sample characteristics of the season clothes x can be determined according to the historical combined data of the birth clothes 1, birth clothes 2, birth clothes 4, birth clothes 6, birth clothes 9, and birth clothes 10. The season wear y is a second sample server, the birth wear 3, the birth wear 5, the birth wear 7 and the birth wear 8 are a plurality of second old servers which are combined to form the season wear y, and similarly, the second sample characteristics of the season wear y can be determined according to the historical combined wear data of the birth wear 3, the birth wear 5, the birth wear 7 and the birth wear 8.
And then constructing a pair-wise based combined feature based on the first sample feature and the second sample feature to obtain a comparison sample feature of the sample server to P. And then, marking corresponding preset scores of each sample server, such as corresponding active scores for corresponding comparison sample characteristics, taking a sample server pair P (a season suit x and a season suit y) as an example, wherein the season suit x is a first sample server, the season suit y is a second sample server, the active scores of the season suit x are 80 scores, the active scores of the season suit y are 85 scores, the higher the active scores are, the better the running effect of the season suit is, and because 85>80 shows that the running effect of the season suit y is compared with the number of the season suit x, the marking information of the comparison sample characteristics of the sample server pair P can be 1, otherwise, the marking information of the comparison sample characteristics of the sample server pair P can be 0, the comparison sample characteristics of each sample server pair are marked, the preset ordering network is trained by adopting the comparison sample characteristics after marking, and a prediction ordering model can be obtained. The preset ordering network may be a neural network such as DNN.
In an embodiment, comparing the first sample feature with the second sample feature to obtain a comparison sample feature of the corresponding sample server pair includes: and dividing the first sample characteristic and the second sample characteristic to obtain a comparison sample characteristic of the corresponding sample server pair.
In this embodiment, the sample attribute data and the sample relationship data of the old server are very much, the corresponding feature vector can reach 108 dimensions, the data flow is very large, when the sample feature is generated, the comparison sample feature can be obtained by dividing the first sample feature and the second sample feature for each sample server pair, so that the data volume of the comparison sample feature is ensured not to be greatly increased, and the calculation efficiency is improved. For example, the feature vector of the first sample feature of a certain sample server pair is 108 dimensions, the feature vector of the second sample feature is 108 dimensions, and in an actual scene, the feature vector and the feature vector can be directly used as the comparison sample feature of the scheme pair, so that the data volume of the comparison sample feature at least comprises 216 dimensions, and the data volume is very large. If the feature vector of the first sample feature is 108 dimensions, the feature vector of the second sample feature is divided by 108 dimensions, and the obtained comparison feature is 108 dimensions, so that compared with the comparison feature, the data dimension of the comparison sample feature is greatly reduced, and the data calculation amount is reduced.
In an embodiment, training a sample feature preset ordering network by using labeled sample feature contrast to generate a prediction ordering model, including: dividing the plurality of sample server pairs according to the merging time, taking the sample server pairs with the merging time before the preset time point as a training set, and taking the sample server pairs with the merging time after the preset time point as a test set. And performing supervised training on a preset ordering network by adopting the labeled sample feature contrast sample features corresponding to the training set, and testing a training result by adopting the labeled sample feature contrast sample features corresponding to the testing set to generate a prediction ordering model.
In this embodiment, a process of supervised training may be adopted to train a model, based on historical fit data, multiple sample servers divide the merged time nodes, a certain preset time node is selected, and the historical season service merged before the time node is assumed to have 700 season servers, so that 40000 groups of supervision data in total are obtained as a training set through the construction of the pair-wise based combined sample features, 10% of the data can be extracted from the training set as verification data, and the rest 90% of the data are used as training data, so as to adjust model parameters, and make the model more accurate. In addition, 400 racing season servers after the preset time node can be selected, and about 8000 groups of data pairs are obtained as a test set after the combination sample characteristics based on pair-wise are constructed. The test set is used for testing the prediction result of the preset sequencing model and assisting in perfecting model training. Therefore, the time node front color history fit data is adopted for training, and the time node rear history fit data is used for testing, so that time division is adopted, space-time crossing is avoided, a model training result is more similar to a real scene, and accuracy is improved.
As shown in fig. 4, a schematic diagram of a predictive model training process according to an embodiment of the application may include the following procedures:
step1 for each sample server in a single sample server pair, features of a single birth service forming the sample server are first extracted, the features encompassing birth service attribute features and relationship features. In addition, a birth service with the number of people ranked ahead, such as a birth service with the number of people ranked ahead four, is selected, a single birth service sample attribute feature and an overall sample attribute feature of the sample server are counted by taking the birth service as a unit, and finally, a feature vector of the overall sample attribute feature and a vector of the single birth service sample attribute feature are spliced to obtain a sample feature of the sample server. Because the pair-wise adopts the comparison characteristic, the sample characteristics to be compared can be directly divided, and finally the comparison sample characteristics of the characterization server pair can be obtained, the characteristic dimension can reach 108 dimensions, and the server characteristic attribute is basically covered. As shown in fig. 4, assuming that the pair of sample servers [ server_x, server_y ], the first sample server server_x is the season service x formed after the service in the historical service data, and the second sample server server_y is the season service y formed after the service in the historical service data, the birth service sample attribute features of the sample server may include the following dimensions: player activity data, recharge attribute data, player achievement data, etc., the birth service sample relationship features may include: characteristic dimensions such as combat relation, alliance relation, friendly relation and the like. After the sample server performs feature extraction on the [ server_x, server_y ], a data dimension obtained by adding the first sample feature and the second sample feature is dim=108×2, and feature comparison processing can be performed on the first sample feature and the second sample feature, for example, the first sample feature and the second sample feature are divided, so that a final comparison sample feature data dimension is dim=108.
step2, setting a label of the characteristic of the comparison sample, combining the season servers which are taken in the same section in pairs according to the historical fit data, and assuming that the sample server pair [ server_x, server_y ] is a season garment x formed after the combination in the historical fit data, the server_y is a season garment y formed after the combination in the historical fit data, the server_x_score is a score of the season garment x in two dimensions of the activity and/or the situation, and the server_y_score is a score of the season garment y in two dimensions of the activity and/or the situation, if the server_x_score is < server_y_score, the label of the characteristic of the comparison sample is set to be 1, otherwise, the label is set to be 0, so that the comparison label (0, 1) can be obtained. Selecting a certain time node, assuming that the historical season service combined before the time node has 700 season servers, obtaining 40000 groups of total supervision data as a training set by constructing the pair-wise based combined sample characteristics, extracting 10% of data from the training set as verification data and the other 90% as training data, so as to adjust model parameters, and enabling the model to be more accurate. In addition, 400 racing season servers after the preset time node can be selected, and about 8000 groups of data pairs are obtained as a test set after the combination sample characteristics based on pair-wise are constructed.
step3. The model may adopt a DNN network structure, the network structure includes a full connection layer, and the specific full connection layer may include Add & Norm (Add and Normalization, normalization layer) and hidden layer (hidden layer), where the dimensions of the full connection layer are 108×16, 16×8,8×2, and batch normalization (batch normalization) layers may be added to accelerate the convergence speed of training and keep the training stable. After training, the model may output a predictive label (predicted_label).
In the experiment, the accuracy of the trained sequencing model on the verification set reaches 91%, and the accuracy on the test set reaches 90%. It can be seen that the model performs better than the human estimated level (about 70%) with a certain business training, and can provide more reliable evaluation results for the fit.
According to the server merging result prediction method, for the problems that the existing server has less history merging data and very sparse data, a conventional point-wise prediction scheme cannot be supported, and the like, the advantages and disadvantages of different server merging schemes are obtained by constructing a pair-wise based problem model and comparing the angles of features, so that the problem that training data is insufficient is effectively solved. For complex game mechanism, a plurality of factors influencing the combined service result are numerous, the game ecology is always changed along with time, and according to the existing historical data, the problem of the law of the game is difficult to learn through simple technical schemes such as statistics, similarity calculation and the like. Compared with the traditional personal experience prediction based on operators, the scheme is more controllable for the state after the combination, and has the main advantages that: (1) From game player's angle, ensured the ecological balance nature of recreation better, promoted player experience. (2) From the angle of developer, reduce ecological operation personnel threshold, promote operating efficiency, reduce the resource input, effectively promote the user and remain and the player is active etc. operation index.
Please refer to fig. 5, which is an embodiment of a method for predicting a server merge result according to the present application, wherein the method can be executed by the electronic device 1 shown in fig. 1 and can be applied to the application scenario of the server merge result prediction shown in fig. 2, so as to effectively enhance data of a small amount of historical merge data, automatically predict the effect of multiple merge schemes by using limited historical merge data, and improve the prediction accuracy of the server merge scheme. In this embodiment, taking the terminal 220 as an executing terminal as an example, the method includes the following steps:
step 501: according to the historical service proposal, a plurality of sample server pairs and preset scores of each sample server are obtained, the sample server pairs comprise a first sample server and a second sample server, the first sample server is obtained by combining a plurality of first old servers, and the second sample server is obtained by combining a plurality of second old servers. See the description of step 304 in the previous embodiments for details.
Step 502: for a single sample server pair, determining first sample features of a corresponding first sample server according to historical compliance program data of a corresponding plurality of first old servers, and determining second sample features of a second sample server according to historical compliance data of a corresponding plurality of second old servers. See the description of step 304 in the previous embodiments for details.
Step 503: and dividing the first sample characteristic and the second sample characteristic to obtain a comparison sample characteristic of the corresponding sample server pair. See the description of step 304 in the previous embodiments for details.
Step 504: and labeling the comparison sample characteristics according to the first preset score of the first sample server and the second preset score of the second sample server, and obtaining labeled comparison sample characteristics of the corresponding sample server pairs. See the description of step 304 in the previous embodiments for details.
Step 505: dividing the plurality of sample server pairs according to the merging time, taking the sample server pairs with the merging time before the preset time point as a training set, and taking the sample server pairs with the merging time after the preset time point as a test set. See the description of step 304 in the previous embodiments for details.
Step 506: and performing supervised training on a preset ordering network by adopting the labeled comparison sample features corresponding to the training set, and testing training results by adopting the labeled comparison sample features corresponding to the testing set to generate a prediction ordering model. See the description of step 304 in the previous embodiments for details.
Step 507: and obtaining a plurality of to-be-predicted combined service schemes, wherein one combined service scheme comprises a plurality of old servers to be combined. See the description of step 301 for details in the previous embodiments.
Step 508: and respectively acquiring the current operation data of a plurality of old servers according to a single compliance scheme. The current operational data includes attribute data and relationship data corresponding to the old server. See the description of step 302 in the previous embodiments for details.
Step 509: and determining the integral attribute characteristics of the new server formed by the corresponding single compliance solution according to the attribute data of each old server. See the description of step 302 in the previous embodiments for details.
Step 510: and obtaining the prediction characteristics of the new server according to the attribute data, the relation data and the overall attribute characteristics of the new server of each old server. See the description of step 302 in the previous embodiments for details.
Step 511: combining the multiple concurrent taking schemes two by two to obtain at least one scheme pair, wherein the scheme pair comprises a first concurrent taking scheme and a second concurrent taking scheme, the first concurrent taking scheme has a first prediction characteristic, and the second concurrent taking scheme has a second prediction characteristic. And dividing the first prediction feature and the second prediction feature to obtain the contrast feature of the corresponding scheme pair. See the description of step 303 in the previous embodiments for details.
Step 512: and inputting the comparison features of at least one scheme pair into a trained preset sequencing model for identification, and outputting sequencing results of a plurality of combined service schemes. See the description of step 304 in the previous embodiments for details.
The details of the steps of the method for predicting the merging result by the server can be referred to the related descriptions of the above embodiments, and are not repeated here.
Please refer to fig. 6, which is a training method of a predictive ranking model according to an embodiment of the present application, the method may be executed by the electronic device 1 shown in fig. 1, and may be applied to an application scenario of server merging result prediction shown in fig. 2, so as to effectively enhance data of a small amount of historical merging data, and improve the prediction effect of the ranking model by using limited historical merging data, thereby improving the prediction accuracy of the server merging scheme. In this embodiment, taking the terminal 220 as an executing terminal as an example, the method includes the following steps:
step 601: according to a plurality of history service schemes, a plurality of sample server pairs and preset scores of all sample servers are obtained, wherein each sample server pair comprises a first sample server and a second sample server, and each history service scheme comprises a plurality of first old servers combined to form the first sample server and a plurality of second old servers combined to form the second sample server.
Step 602: for a single sample server pair, determining first sample features of a corresponding first sample server according to historical compliance program data of a corresponding plurality of first old servers, and determining second sample features of a second sample server according to historical compliance data of a corresponding plurality of second old servers.
Step 603: and constructing a comparison sample characteristic of the corresponding sample server pair according to the first sample characteristic and the second sample characteristic.
Step 604: and labeling the comparison sample characteristics according to the first preset score of the first sample server and the second preset score of the second sample server, and obtaining labeled comparison sample characteristics of the corresponding sample server pairs.
Step 605: training a preset ordering network by comparing the marked sample characteristics to generate a prediction ordering model.
The details of each step of the above prediction ranking model training method can be referred to the related description of the above model training embodiment, and will not be repeated here.
Please refer to fig. 7, which illustrates a server merge result prediction apparatus 700 according to an embodiment of the present application, which can be applied to the electronic device 1 illustrated in fig. 1 and can be applied to the application scenario of server merge result prediction illustrated in fig. 2, so as to effectively enhance data of a small amount of historical merge data, and automatically predict the effect of multiple merge schemes by using limited historical merge data, thereby improving the prediction accuracy of the server merge scheme. The device comprises: the functional principles of the acquisition module 701, the determination module 702, the construction module 703 and the identification module 704 are as follows:
The obtaining module 701 is configured to obtain a plurality of to-be-predicted compliance schemes, where one compliance scheme includes a plurality of old servers to be combined.
The determining module 702 is configured to determine, for a single compliance solution, a predicted feature of a new server formed by the single compliance solution according to feature data corresponding to a plurality of old servers.
The construction module 703 is configured to combine the multiple concurrent taking schemes two by two to obtain at least one pair of schemes, where the pair of schemes includes a first concurrent taking scheme and a second concurrent taking scheme, the first concurrent taking scheme has a first prediction feature, and the second concurrent taking scheme has a second prediction feature. And constructing contrast features of the corresponding scheme pairs according to the first prediction features and the second prediction features.
The identifying module 704 is configured to input the comparison features of at least one of the solution pairs into a trained preset ranking model for identification, output ranking results of the multiple concurrent solutions, and the preset ranking model is obtained based on the comparison sample features of the multiple historical concurrent solutions.
In one embodiment, the feature data includes current operation data, and the determining module 702 is specifically configured to obtain the current operation data of the plurality of old servers for a single compliance solution. And determining the prediction characteristics of the new server formed corresponding to the single compliance program according to the current operation data.
In one embodiment, the current operational data includes attribute data and relationship data corresponding to the old server, wherein the attribute data is used to characterize player attributes within the corresponding old server. The relationship data is used to characterize the relationship between different old servers in a single compliance solution. And the determining module 702 is configured to determine, according to attribute data of each old server, an overall attribute characteristic of a new server formed corresponding to the single compliance solution. And obtaining the prediction characteristics of the new server according to the attribute data, the relation data and the overall attribute characteristics of the new server of each old server.
In an embodiment, the construction module 703 is configured to divide the first predicted feature and the second predicted feature to obtain a comparison feature of the corresponding scheme pair.
In one embodiment, the method further comprises: the training module is used for training the prediction sequencing model before inputting the contrast characteristic of at least one scheme pair into the trained preset sequencing model for recognition and outputting the sequencing results of a plurality of combined service schemes, and the training module is specifically used for: according to a plurality of history service schemes, a plurality of sample server pairs and preset scores of all sample servers are obtained, wherein each sample server pair comprises a first sample server and a second sample server, and each history service scheme comprises a plurality of first old servers combined to form the first sample server and a plurality of second old servers combined to form the second sample server. For a single sample server pair, determining first sample features of a corresponding first sample server according to historical compliance program data of a corresponding plurality of first old servers, and determining second sample features of a second sample server according to historical compliance data of a corresponding plurality of second old servers. And constructing a comparison sample characteristic of the corresponding sample server pair according to the first sample characteristic and the second sample characteristic. And labeling the comparison sample characteristics according to the first preset score of the first sample server and the second preset score of the second sample server, and obtaining labeled comparison sample characteristics of the corresponding sample server pairs. Training a preset ordering network by comparing the marked sample characteristics to generate a prediction ordering model.
In an embodiment, the training module is specifically configured to divide the first sample feature and the second sample feature to obtain a comparison sample feature of the corresponding sample server pair.
In an embodiment, the training module is further specifically configured to divide the plurality of sample server pairs according to the merging time, use a sample server pair with a merging time before a preset time point as the training set, and use a sample server pair with a merging time after the preset time point as the testing set. And performing supervised training on a preset ordering network by adopting the labeled comparison sample features corresponding to the training set, and testing training results by adopting the labeled comparison sample features corresponding to the testing set to generate a prediction ordering model.
For a detailed description of the above-mentioned server merging result prediction apparatus 700, please refer to the description of the related method steps in the above-mentioned embodiment, the implementation principle and technical effects are similar, and the detailed description of this embodiment is omitted herein.
Fig. 8 is a schematic structural diagram of a cloud device 80 according to an exemplary embodiment of the present application. The cloud device 80 may be used to run the methods provided in any of the embodiments described above. As shown in fig. 8, the cloud device 80 may include: memory 804 and at least one processor 805, one for example in fig. 8.
Memory 804 is used to store computer programs and may be configured to store various other data to support operations on cloud device 80. The memory 804 may be an object store (Object Storage Service, OSS).
The memory 804 may be implemented by any type or combination of volatile or nonvolatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disk.
The processor 805 is coupled to the memory 804, and is configured to execute the computer program in the memory 804, so as to implement the solutions provided by any of the method embodiments, and specific functions and technical effects that can be implemented are not described herein.
Further, as shown in fig. 8, the cloud device further includes: firewall 801, load balancer 802, communication component 806, power component 803, and other components. Only some components are schematically shown in fig. 8, which does not mean that the cloud device only includes the components shown in fig. 8.
In one embodiment, the communication component 806 of fig. 8 is configured to facilitate wired or wireless communication between the device in which the communication component 806 is located and other devices. The device in which the communication component 806 is located can access a wireless network based on a communication standard, such as a WiFi,2G, 3G, 4G, LTE (Long Term Evolution, long term evolution, LTE for short), 5G, or a combination thereof. In one exemplary embodiment, the communication component 806 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component 806 further includes a near field communication (Near Field Communication, NFC for short) module to facilitate short range communications. For example, the NFC module may be implemented based on radio frequency identification (Radio Frequency Identification, RFID) technology, infrared data association (Infrared Data Association, irDA) technology, ultra Wide Band (UWB) technology, bluetooth (BT) technology, and other technologies.
In one embodiment, the power component 803 of fig. 8 provides power to various components of the device in which the power component 803 is located. The power components 803 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the devices in which the power components reside.
The embodiment of the application also provides a computer readable storage medium, wherein computer executable instructions are stored in the computer readable storage medium, and when the processor executes the computer executable instructions, the method of any of the previous embodiments is realized.
Embodiments of the present application also provide a computer program product comprising a computer program which, when executed by a processor, implements the method of any of the preceding embodiments.
In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described device embodiments are merely illustrative, e.g., the division of modules is merely a logical function division, and there may be additional divisions of actual implementation, e.g., multiple modules may be combined or integrated into another system, or some features may be omitted or not performed.
The integrated modules, which are implemented in the form of software functional modules, may be stored in a computer readable storage medium. The software functional modules described above are stored in a storage medium and include instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or processor to perform some of the steps of the methods of the various embodiments of the application.
It should be appreciated that the processor may be a central processing unit (Central Processing Unit, CPU for short), other general purpose processors, digital signal processor (Digital Signal Processor, DSP for short), application specific integrated circuit (Application Specific Integrated Circuit, ASIC for short), etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the present application may be embodied directly in a hardware processor for execution, or in a combination of hardware and software modules in a processor for execution. The memory may include a high-speed RAM (Random Access Memory ) memory, and may further include a nonvolatile memory NVM (Nonvolatile memory, abbreviated as NVM), such as at least one magnetic disk memory, and may also be a U-disk, a removable hard disk, a read-only memory, a magnetic disk, or an optical disk.
The storage medium may be implemented by any type of volatile or non-volatile Memory device or combination thereof, such as Static Random-Access Memory (SRAM), electrically erasable programmable Read-Only Memory (Electrically Erasable Programmable Read Only Memory, EEPROM), erasable programmable Read-Only Memory (Erasable Programmable Read-Only Memory, EPROM), programmable Read-Only Memory (Programmable Read-Only Memory, PROM), read-Only Memory (ROM), magnetic Memory, flash Memory, magnetic disk, or optical disk. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.
An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an application specific integrated circuit (Application Specific Integrated Circuits, ASIC for short). It is also possible that the processor and the storage medium reside as discrete components in an electronic device or a master device.
It should be noted that, in this document, 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 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 one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The foregoing embodiment numbers of the present application are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising several instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method of the embodiments of the present application.
In the technical scheme of the application, the related information such as user data and the like is collected, stored, used, processed, transmitted, provided, disclosed and the like, which are all in accordance with the regulations of related laws and regulations and do not violate the popular public order.
The foregoing description is only of the preferred embodiments of the present application, and is not intended to limit the scope of the application, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.