CN115779444B - Cloud game data safety protection method and server applied to artificial intelligence - Google Patents

Abstract

The embodiment of the invention optimizes basic local risk detection description of cloud game player behavior data under a local characteristic environment based on basic global risk detection description of a node aggregation set under a global characteristic environment, and the accuracy and reliability of the optimized local risk detection description are ensured because a plurality of behavior characteristic nodes corresponding to one node aggregation set under the global characteristic environment can exist, and the cloud game player behavior data corresponding to each behavior characteristic node can exist. In addition, through the risk detection processing thought, the risk detection of the cloud game player behavior data can be effectively realized, so that the normal use of the optimized local risk detection description of each group of cloud game player behavior data in the subsequent game data protection process is ensured.

Description

Cloud game data safety protection method and server applied to artificial intelligence

Technical Field

The invention relates to the technical field of artificial intelligence and cloud games, in particular to a cloud game data safety protection method and a server applied to artificial intelligence.

Background

Cloud gaming (close gaming) refers to deploying all games on a cloud server, performing game rendering and picture processing through the cloud server, and then transmitting to a game client through a network. Accordingly, the game client only needs to upload relevant game operation data to the cloud server, so that the normal running of the cloud game can be realized. In order to enable a game player to have a good game experience, the problem of data security protection of the game player during the game operation is not neglected. When data security protection processing is carried out on cloud games, game risk detection is an important circle, however, when the game risk detection is carried out in the related technology, the detection efficiency and quality of the related technology are difficult to meet the actual requirements.

Disclosure of Invention

The invention provides a cloud game data safety protection method and a server applied to artificial intelligence.

The first aspect is a cloud game data security protection method applied to artificial intelligence, and applied to an artificial intelligence server, the method at least comprises: determining a player behavior feature library generated from n sets of cloud player behavior data, and a base local risk detection description for each set of cloud player behavior data; wherein n is a positive integer; sorting each behavior feature node covered by the player behavior feature library into a plurality of node aggregation sets; carrying out global risk detection on each node cluster obtained through the overall management through the basic local risk detection description to obtain basic global risk detection description of each node cluster; and acquiring a local risk detection description optimized for the completion of the cloud game player behavior data of each group by combining the basic global risk detection description of each node aggregation set.

In this way, on the basis of performing clustering arrangement of node aggregation sets on a player behavior feature library generated based on cloud game player behavior data, global risk detection description can be performed on each node aggregation set obtained through overall arrangement based on basic local risk detection description corresponding to each cloud game player behavior data, corresponding basic global risk detection description is determined, and based on basic global risk detection description, the optimized local risk detection description for each cloud game player behavior data can be obtained, in other words, the embodiment of the invention optimizes basic local risk detection description of cloud game player behavior data in a local feature environment based on basic global risk detection description of the node aggregation set in the global feature environment, and as a plurality of behavior feature nodes corresponding to one node aggregation set in the global feature environment can exist, cloud game player behavior data corresponding to each behavior feature node can exist, the node aggregation set can be subjected to packaged risk detection processing through the technology thought, so that the accuracy and the reliability of the optimized local risk detection description can be obtained, in addition, the high-efficiency of the cloud game player behavior data can be guaranteed through the detection thought processing, and the normal risk detection of each cloud game player behavior data can be achieved, and the game player behavior data can be guaranteed to be detected in a normal mode.

For an illustrative embodiment, the sorting each behavior feature node covered by the player behavior feature library into a plurality of node aggregation sets includes: randomly determining target number of behavior feature nodes from each behavior feature node covered by the player behavior feature library, and taking the key feature information of each determined behavior feature node as the reference feature information of the node aggregation set to be clustered and arranged; determining the key feature information of each behavior feature node in each behavior feature node, sequentially comparing the key feature information with each reference feature information in a quantitative comparison mode, and sorting the behavior feature nodes into a node aggregation set to which the corresponding reference feature information with the quantitative comparison result meeting the specified requirements belongs; and obtaining current reference characteristic information corresponding to each node aggregation set, and combining the current reference characteristic information, and jumping to the step of arranging the node aggregation sets until the node aggregation sets accord with the clustering evaluation indexes.

Therefore, the arrangement of the node aggregation sets not lower than one round can be performed based on the key characteristic information of the behavior characteristic nodes and the reference characteristic information of each node aggregation set. And determining the current reference characteristic information for the node cluster set every time of finishing, thereby finishing the next round. In view of the fact that the behavior feature nodes arranged into one node aggregation set have relevance to a certain extent, basic global risk detection description of the node aggregation set can be determined by means of risk detection content corresponding to the behavior feature nodes of one node aggregation set, so that the accuracy of the detection content can be guaranteed, and the accuracy of the optimized local risk detection description can be further guaranteed.

For an exemplary embodiment, the obtaining the current reference feature information corresponding to each node cluster set includes: and combining key characteristic information of each behavior characteristic node covered by each node cluster set to obtain depolarization characteristic information of each behavior characteristic node, and taking the depolarization characteristic information as current reference characteristic information corresponding to the node cluster set.

For one illustrative embodiment, the key feature information includes a key indicative feature and a target differential description, and the reference feature information includes a reference indicative feature and a reference differential description; the determining the quantitative comparison result between the key feature information of each behavior feature node in each behavior feature node and each reference feature information in turn comprises the following steps: determining a first quantitative comparison result between the behavior feature node and a node cluster to which each piece of reference feature information belongs by combining key indication type features of the behavior feature node and reference indication type features of each piece of reference feature information aiming at each behavior feature node in the behavior feature nodes, and determining a second quantitative comparison result between the behavior feature node and the node cluster to which each piece of reference feature information belongs by combining target differential description of the behavior feature node and reference differential description of each piece of reference feature information; and determining the key characteristic information of the behavior characteristic node through the first quantitative comparison result and the importance index matched with the first quantitative comparison result and the second quantitative comparison result and the importance index matched with the second quantitative comparison result, and the quantitative comparison result between the key characteristic information and each piece of reference characteristic information in sequence.

In this way, in view of the fact that the indication type features and the differential descriptions of the behavior feature nodes are different in interference degree to the node cluster set arrangement, the above key feature information about the behavior feature nodes can be implemented according to the thought of global processing, and the determination of the quantitative comparison result between the key feature information and each piece of reference feature information can be performed sequentially, so that the technical scheme can be applied to different scenes as much as possible.

For an illustrative embodiment, the performing global risk detection on each node cluster obtained by the overall process through the basic local risk detection description to obtain a basic global risk detection description of each node cluster includes: for each behavior feature node, determining a mapping data set corresponding to the behavior feature node in n groups of cloud game player behavior data based on the corresponding situation between a first continuous space corresponding to the player behavior feature library and a second continuous space corresponding to the cloud game player behavior data; determining a basic global risk detection description of the behavior feature node in combination with the basic local risk detection description of the mapping data set corresponding to the behavior feature node in the n groups of cloud game player behavior data; and aiming at each node aggregation set obtained by arrangement, combining the basic global risk detection description of each behavior characteristic node covered by the node aggregation set, and determining the basic global risk detection description of the node aggregation set.

In this way, the basic global risk detection description of the corresponding behavior feature nodes can be determined together based on the basic local risk detection description of the behavior data of the cloud game player, and the basic global risk detection description of the node cluster set can be determined based on the basic global risk detection description of a plurality of behavior feature nodes covered by one node cluster set.

For an illustrative embodiment, said determining a base global risk detection description for the behavior feature node in combination with said base local risk detection description for the mapping dataset corresponding to the behavior feature node in the n sets of cloud game player behavior data comprises: for each behavior feature node, determining keyword statistical results corresponding to various basic local risk detection descriptions by combining basic local risk detection descriptions of corresponding mapping data sets of the behavior feature node in the n groups of cloud game player behavior data; and combining keyword statistical results corresponding to various basic local risk detection descriptions to determine basic global risk detection descriptions of the behavior feature nodes.

For an illustrative embodiment, the determining the basic global risk detection description of the behavior feature node according to the keyword statistics corresponding to each type of basic local risk detection description includes: determining a basic local risk detection description of which the keyword statistical result meets a set statistical condition as a basic global risk detection description of the behavior feature node; the determining, for each node cluster obtained by the sorting, a basic global risk detection description of the node cluster in combination with a basic global risk detection description of each behavior feature node covered by the node cluster includes: combining basic global risk detection descriptions of each behavior feature node covered by each node aggregation set, and determining statistical values of the behavior feature nodes corresponding to various basic global risk detection descriptions; and determining the basic global risk detection description of the node aggregation set according to the basic global risk detection description of the class with the maximum statistical value of the corresponding behavior feature nodes.

For an illustrative embodiment, the determining the basic global risk detection description of the behavior feature node according to the keyword statistics corresponding to each type of basic local risk detection description includes: carrying out quantitative analysis processing on keyword statistical results and keyword summation results corresponding to various basic local risk detection descriptions, and determining that the behavior feature nodes are matched with quantitative prediction results of each type of basic local risk detection description; determining basic global risk detection description of the behavior feature node according to quantitative prediction results matched with various basic local risk detection descriptions; the determining, for each node cluster obtained by the sorting, a basic global risk detection description of the node cluster in combination with a basic global risk detection description of each behavior feature node covered by the node cluster includes: for each node cluster set obtained by arrangement, combining a quantized prediction result which corresponds to each behavior feature node covered by the node cluster set and is matched with each type of basic local risk detection description, and an importance index which is previously set for the behavior feature node and corresponds to each type of basic local risk detection description, and determining the quantized prediction result which corresponds to each type of basic local risk detection description; and combining the quantized prediction results of the node cluster set matched with each type of basic local risk detection description to determine the basic global risk detection description of the node cluster set.

For one illustrative embodiment, upon determining that any node cluster matches a quantized prediction result of the multi-class basis local risk detection description that exceeds a specified prediction result, the method further comprises: for any node cluster set, the node cluster set is arranged into a plurality of node cluster subsets;

the global risk detection is performed on each node cluster obtained through the basic local risk detection description, and basic global risk detection description of each node cluster is obtained, including:

carrying out global risk detection on each node cluster subset obtained through the overall management through the basic local risk detection description to obtain basic global risk detection description of each node cluster subset;

the obtaining, in combination with the global risk detection description of each node aggregation set, a local risk detection description optimized for completion of the cloud game player behavior data of each group includes: and obtaining the optimized local risk detection description aiming at the behavior data of each cloud game player by the basic global risk detection description of each node cluster subset.

Therefore, on the basis of determining that one node cluster set has multiple risk detection possibilities, the node cluster subsets can be sorted firstly, and then the risk detection is carried out according to the sorted node cluster subsets, so that the accuracy and the reliability of the risk detection can be obviously improved.

For an illustrative embodiment, said obtaining, in conjunction with said base global risk detection description for each said node cluster set, a locally risk detection description optimized for completion of said each set of cloud game player behavior data comprises: determining a global risk detection description for completing optimization; the global risk detection description after optimization is active global risk detection description after optimization is carried out on basic global risk detection description of at least a part of node cluster sets in the plurality of node cluster sets; and obtaining the optimized local risk detection description aiming at the behavior data of each cloud game player through the optimized global risk detection description and the basic global risk detection description of the residual node aggregation set which does not finish the active optimization.

In this way, in view of the interference of the accuracy of the optimization completion local risk detection description on the later correlation after the active optimization, the optimization completion local risk detection description for each cloud game player behavior data group can be obtained based on the optimization completion global risk detection description of the active optimization of at least a part of the node aggregation set, so that the accuracy and the reliability of risk detection can be further improved significantly.

For an illustrative embodiment, said obtaining, by said base global risk detection description of each said node cluster set, a locally risk detection description optimized for completion of said each set of cloud player behavior data comprises: combining the basic global risk detection description of each node cluster set to obtain global prediction condition offset corresponding to each node cluster set in the risk detection dimension; obtaining quantized difference offset corresponding to each node cluster set in the behavior trend dimension according to quantized difference data between two node cluster sets which are random and have association relations and are not matched with the same global risk detection description; determining a risk detection optimization model through the global prediction situation offset and the quantitative difference offset; determining an adjusted global risk detection description determined for each node cluster on the basis of ensuring that the output variable of the risk detection optimization model meets a set detection condition; and acquiring a local risk detection description optimized for the completion of the cloud game player behavior data of each group by combining the adjusted global risk detection description for each node aggregation set.

In this way, optimization of local risk detection description on each cloud game player behavior data can be performed by combining global prediction condition offset corresponding to each node cluster set and quantization difference offset pointed by any node cluster set with association relation, and the optimization is automatically realized, so that the labeling efficiency can be further improved.

For an illustrative embodiment, said obtaining, in conjunction with said base global risk detection description for each said node cluster set, a locally risk detection description optimized for completion of said each set of cloud game player behavior data comprises: for each node aggregation set, determining cloud game player behavior data corresponding to each behavior feature node covered by the node aggregation set from the n groups of cloud game player behavior data based on the corresponding situation between a first continuous space corresponding to the player behavior feature library and a second continuous space corresponding to the cloud game player behavior data; and aiming at the behavior data of each cloud game player, combining the basic global risk detection description of the node aggregation set corresponding to the behavior feature nodes corresponding to the behavior data of the cloud game player to acquire the local risk detection description of the completion optimization aiming at the behavior data of the cloud game player.

For an illustrative embodiment, the base local risk detection is described as a result of risk detection of the each set of cloud game player behavior data by means of a base game risk detection model, and the obtaining of the optimized local risk detection description for the completion of the each set of cloud game player behavior data further comprises, after: and configuring the basic game risk detection model through the optimized local risk detection description of each group of cloud game player behavior data to obtain a configured game risk detection model.

A second aspect is an artificial intelligence server comprising a memory and a processor; the memory is coupled to the processor; the memory is used for storing computer program codes, and the computer program codes comprise computer instructions; wherein the computer instructions, when executed by the processor, cause the artificial intelligence server to perform the method of the first aspect.

Drawings

Fig. 1 is a schematic flow chart of a cloud game data security protection method applied to artificial intelligence according to an embodiment of the present invention.

Fig. 2 is a block diagram of a cloud game data security device applied to artificial intelligence according to an embodiment of the present invention.

Detailed Description

Hereinafter, the terms "first," "second," and "third," etc. are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first", "a second", or "a third", etc., may explicitly or implicitly include one or more such feature.

FIG. 1 illustrates a flow diagram of a cloud gaming data security method applied to artificial intelligence, which may be implemented by an artificial intelligence server, and the artificial intelligence server may include a memory and a processor; the memory is coupled to the processor; the memory is used for storing computer program codes, and the computer program codes comprise computer instructions; wherein the computer instructions, when executed by the processor, cause the artificial intelligence server to perform the technical scheme described in the following steps.

Step 101, determining a player behavior feature library generated according to n groups of cloud game player behavior data, and a basic local risk detection description of each group of cloud game player behavior data.

For purposes of embodiments of the present invention, cloud game player behavior data may be understood as a series of records that a cloud game player produces during a game play operation. Such as: a key mouse operation record, game session data, game auxiliary software operation data, and the like.

Step 102, each behavior feature node covered by the player behavior feature library is arranged into a plurality of node aggregation sets.

And 103, carrying out global risk detection on each node cluster obtained through the whole process based on the basic local risk detection description, and obtaining basic global risk detection description of each node cluster.

Step 104, combining the basic global risk detection description of each node aggregation set, and obtaining the local risk detection description of the completion optimization of the behavior data of each cloud game player.

It will be appreciated that, for ease of understanding the cloud game data security protection method applied to artificial intelligence shown in the embodiments of the present invention, an exemplary implementation environment of the cloud game data security protection method may be adaptively described. It can be appreciated that the cloud game data security protection method applied to artificial intelligence can be applied to a game environment where risk detection is performed by Yu Renyi cloud game interaction items, for example: the risk detection of the game recharging fraud in the man-machine mode can be understood as the risk detection of the fishing advertisement in the game playing process of the player, and the risk detection of other game interaction matters (such as Trojan horse detection and the like), and the embodiment of the invention is not limited further.

While the related detection concepts focus on carpet type detection, the carpet type risk detection concepts may be accomplished through active type detection. In view of the fact that many reference game player behavior data that have been detected are typically required in performing the model configuration, this may result in unnecessary resource waste. It can be appreciated that, in order to improve the above existing technical problems, the embodiment of the present invention shows a more efficient and intelligent risk detection technology, which can perform risk detection in a global feature environment, and then map a risk detection description in the global feature environment to a local feature environment based on a corresponding situation between the global feature environment and the local feature environment, so that the quality of risk detection can be significantly improved to a certain extent, unnecessary resource waste is reduced, and timeliness of risk detection is improved.

To enable risk detection in a global feature environment, a library of player behavior features generated based on one or more sets of cloud game player behavior data and a base local risk detection description for each set of cloud game player behavior data may be first determined. Wherein, the player behavior feature library may be understood as being determined based on the creation of the integrated knowledge base based on cloud game player behavior data. Further, the integrated knowledge base creation may generally include the following links: knowledge unit positioning, connection relation determination and knowledge base construction, so that the player behavior feature base can be orderly and completely determined.

Furthermore, the basic local risk detection description about cloud game player behavior data in the embodiment of the present invention may be understood as being detected by means of a basic game risk detection model. Based on the basic local risk detection description of the behavior data of the cloud game player and the feature correspondence situation between the global feature environment and the global feature environment, global risk detection (for example, multi-dimensional risk detection can be understood) of the behavior feature nodes covered by the behavior feature library of the player can be realized. Further, the basic game risk detection model may be understood as being configured by using the reference player behavior data of completed risk detection recorded in the historical risk information set, and the configured reference player behavior data of completed risk detection may be understood as a mapping condition (such as a mapping relationship or a corresponding relationship) between the reference player behavior data of completed risk detection and authentication annotations (such as labeling information).

For the embodiment of the present invention, the player behavior feature library (for example, may be recorded by a knowledge base or a knowledge graph form) may be understood as being obtained by combining behavior feature nodes, in other words, the behavior feature nodes may be understood as Min units (minimum entity/knowledge unit) of the player behavior feature library. In practical implementation, in view of the fact that behavior categories of different behavior feature nodes can be understood to be the same or different, further, the node aggregation set can be arranged according to the behavior feature nodes of the behavior category, unified risk detection is conveniently carried out on the behavior feature nodes belonging to the same behavior category in the node aggregation set, and therefore the quality of risk detection can be remarkably improved on the basis of guaranteeing the accuracy of risk detection.

The basic global risk detection description for a node aggregation set in embodiments of the present invention may be understood as being determined based on basic local risk detection descriptions for cloud game player behavior data. The method mainly aims at the mapping situation between one node aggregation set and a plurality of behavior feature nodes, and the mapping situation between one behavior feature node and a plurality of groups of cloud game player behavior data, so that on the basis of determining the basic local risk detection description of the cloud game player behavior data, the basic global risk detection description of the node aggregation set can be determined on the basis of the mapping situation.

For the embodiment of the invention, on the basis of obtaining the basic global risk detection description of each node aggregation set, the optimized local risk detection description of all cloud game player behavior data can be determined without repetition based on the feature correspondence condition of the global feature environment and the local feature environment, and compared with the thought of annotation one by one in the local feature environment, the quality and the efficiency of detection can be obviously improved.

For embodiments of the present invention, the base game risk detection model may be configured based on a partially risk detection description of the completion optimization of cloud game player behavior data. The accuracy of the optimized local risk detection description is more accurate, so that the game risk detection model is more comprehensive in detection, and further the game risk detection model is convenient to use in various environments in the later period.

For an illustrative embodiment, considering that the arrangement of the node aggregation set plays a very important role in the later risk detection, the idea of the overall node aggregation set may be further described. The collation thread for the node collection may illustratively include what is recorded in steps 1021-1023.

Step 1021, randomly determining target number of behavior feature nodes from each behavior feature node covered by the player behavior feature library, and taking the key feature information of each determined behavior feature node as the reference feature information of the node aggregation set to be clustered and arranged.

For example, the reference feature information may be understood as basic feature information.

Step 1022, obtaining key feature information of each behavior feature node in each behavior feature node, comparing the key feature information with each reference feature information in sequence, and sorting the behavior feature nodes to a node aggregation set to which the corresponding reference feature information with the quantized comparison result meeting the specified requirement belongs.

For example, reference feature information for which the quantized comparison result meets the specified requirement may be understood as reference feature information for which the comparison result is minimum or lowest.

Step 1023, obtaining current reference characteristic information corresponding to each node aggregation set, and combining the current reference characteristic information, and jumping to a step of sorting the node aggregation sets until the clustering evaluation indexes are met.

For example, the current reference feature information may be understood as the latest reference feature information acquired. The cluster evaluation index may be understood as conforming to the behavior feature node arrangement condition, wherein arrangement may be a division process or a classification process.

For embodiments of the present invention, key feature information (which may be understood as the target vector of the smallest entity of the player behavior feature library) includes key-indicative features and/or target-differentiated descriptions, and the descriptions of behavior feature nodes may be reflected from different levels. The descriptions of key indicator features and target differencing may be understood as being determined based on the results of the integrated knowledge base creation.

The embodiment of the invention can carry out a round of arrangement step or not based on the quantitative comparison result between the reference characteristic information of the node aggregation set and the key characteristic information of the behavior characteristic nodes until the clustering evaluation index is met, and a final arrangement result can be obtained.

It can be appreciated that after finishing each round, the current reference feature information corresponding to the node cluster set can be determined, and the next round of finishing is performed based on the current reference feature information. The reference feature information about the node cluster set may be understood as depolarization feature information determined by key feature information of each behavior feature node of the node cluster set, in other words, depolarization results may be understood as determination of key feature information of all behavior feature nodes in the node cluster set.

The clustering evaluation index in the embodiment of the invention can be understood as that for a node cluster set, the reference characteristic information calculated by a plurality of rounds tends to be stable, the sorting round number accords with the set round number, and the clustering evaluation index can be understood as other clustering evaluation indexes, and the embodiment of the invention is not limited to the above.

In view of the fact that the indicated type of feature and the differential description are not identical in terms of interference levels to the arrangement of the node clusters, two quantitative comparison results can be determined according to the two features, and the quantitative comparison results can be determined by adopting a global summation concept, and the method can comprise steps 10221 and 10222.

Step 10221, determining a first quantitative comparison result between the behavior feature node and the node cluster to which each piece of reference feature information belongs by combining the key indication type feature of the behavior feature node and the reference indication type feature of each piece of reference feature information, and determining a second quantitative comparison result between the behavior feature node and the node cluster to which each piece of reference feature information belongs by combining the target differential description of the behavior feature node and the reference differential description of each piece of reference feature information.

Step 10222, determining key feature information of the behavior feature node based on the first quantized comparison result and the importance index matched with the first quantized comparison result, and the second quantized comparison result and the importance index matched with the second quantized comparison result, and the quantized comparison result between the key feature information and each piece of reference feature information in sequence.

For the embodiment of the present invention, the importance index (weight) corresponding to the first quantization comparison result (such as the feature distance, the euclidean distance, the cosine distance, etc.) and the importance index corresponding to the second quantization comparison result may be set according to different application scenarios. Such as: in some cases, the dissimilarity between the differential description and the reference differential description plays a relatively obvious auxiliary role in the arrangement of the node aggregation sets, so that importance indexes with higher credibility can be arranged.

The arrangement method provided by the embodiment of the invention aims at enabling the behavior feature nodes of the same behavior class to correspond to the same node aggregation set, and the behavior feature nodes of different behavior classes to correspond to different node aggregation sets.

For the embodiment of the present invention, the basic global risk detection description for the node aggregation set is obtained based on the basic local risk detection description corresponding to the cloud game player behavior data, which may exemplarily include the contents recorded in steps 1031-1033.

Step 1031, for each behavior feature node, determining a mapping data set corresponding to the behavior feature node in the n groups of cloud game player behavior data based on the correspondence between the first continuous space corresponding to the player behavior feature library and the second continuous space corresponding to the cloud game player behavior data.

Step 1032, determining a basic global risk detection description of the behavior feature node in combination with the basic local risk detection description of the corresponding mapping dataset of the behavior feature node in the n groups of cloud game player behavior data.

Step 1033, for each node cluster obtained by sorting, determining a basic global risk detection description of the node cluster by combining the basic global risk detection description of each behavior feature node covered by the node cluster.

For the embodiment of the present invention, in view of the correspondence between the first continuous space corresponding to the player behavior feature library and the second continuous space corresponding to the cloud game player behavior data, for each behavior feature node, a mapping data set corresponding to the behavior feature node in each set of cloud game player behavior data may be determined first. The continuous space may be a reference space commonly used in machine learning and artificial intelligence, such as: feature space or map space.

It is understood that the mapping data set may be understood as a summary of each feature unit obtained by mapping data such as a sample space of behavior feature nodes, a player behavior state, a game network environment state, and the like to cloud game player behavior data. In view of the fact that the basic local risk detection descriptions of the cloud player behavior data of each group are detected in advance, the basic local risk detection descriptions of the mapping data sets corresponding to the cloud player behavior data can be determined.

On the basis of determining basic local risk detection descriptions of a mapping data set aiming at the behavior feature nodes, keyword statistics results (such as label numbers) corresponding to various basic local risk detection descriptions can be determined, in other words, statistics of one keyword corresponding to the behavior feature nodes can be obtained, dimensionless processing (normalization processing) is carried out on the statistics of each keyword, behavior category possibility of the behavior feature nodes corresponding to one keyword can be obtained, further, the basic local risk detection description with highest behavior category possibility can be selected, in other words, the keyword statistics results meet the set statistics conditions can be used as basic global risk detection description of the behavior feature nodes.

The node aggregation sets are formed by combining behavior feature nodes, and according to basic global risk detection description of each behavior feature node covered by the node clustering sets, statistics values of the behavior feature nodes corresponding to various basic global risk detection descriptions are determined, and the higher the statistics values of the behavior feature nodes, the higher the probability of the basic global risk detection description corresponding to the node aggregation sets. And selecting the basic global risk detection description with the maximum statistical value of the corresponding behavior feature nodes as the basic global risk detection description of the node aggregation set.

In addition, the embodiment of the invention can also determine the basic global risk detection description of the node cluster according to the following operation.

And 201, carrying out quantitative analysis processing (for example, ratio calculation can be carried out) on keyword statistics results and keyword summation results corresponding to various basic local risk detection descriptions, and determining that the behavior characteristic node is matched with quantitative prediction results (for example, probability values) of each type of basic local risk detection description.

Step 202, determining a basic global risk detection description of the behavior feature node according to quantitative prediction results matched with various basic local risk detection descriptions.

Step 203, for each node cluster set obtained by sorting, determining, in combination with a quantized prediction result corresponding to each behavior feature node covered by the node cluster set and matched with each type of basic local risk detection description, and an importance index corresponding to each type of basic local risk detection description set in advance for the behavior feature node, a quantized prediction result corresponding to each type of basic local risk detection description, where the node cluster set is matched with the node cluster set.

And 204, combining the quantized prediction results of the node cluster set matched with each type of basic local risk detection description to determine the basic global risk detection description of the node cluster set.

For the embodiment of the invention, the determination of the quantitative prediction result matched with each type of basic local risk detection description can be firstly carried out on the behavior feature nodes, and then the determination of the basic global risk detection description aiming at each node aggregation set is realized according to the behavior feature nodes covered by one node aggregation set.

Further, the quantitative predictions regarding a behavior feature node matching each type of underlying local risk detection description are determined based on the calculation (e.g., ratio) of keyword statistics. The hit statistics (corresponding to keyword statistics) of each set of basic local risk detection descriptions determined by a behavior feature node in the process of matching the corresponding mapping data set in each set of cloud game player behavior data can be recorded, all hit statistics (corresponding to keyword summation results) can be recorded, and the quantized prediction results can be determined through the duty ratio of the two statistics.

In addition, under the process of determining the likelihood for a node cluster set, the method can be completed by combining the importance index which is previously set by each behavior feature node covered by the node cluster set and corresponds to each type of basic local risk detection description.

In practical implementation, the importance index can be understood to be determined by combining the activity hotness of the behavior feature nodes, the behavior feature nodes with larger activity hotness can correspondingly set larger importance indexes, and the behavior feature nodes with smaller activity hotness can correspondingly set smaller importance indexes, which is mainly because the behavior feature nodes with large activity hotness generally have more important influence in the clustering process. It will be appreciated that by global integration (weighted summation), it can be determined that a node cluster set matches the quantized prediction results of each type of base local risk detection description, and that a base global risk detection description of a node cluster set can be regarded as a prediction case.

In practical implementation, a problem that risk detection is difficult to directly perform on the node aggregation set occurs to a certain extent. In order to solve the above problems, the embodiment of the present invention may further sort the node cluster set, and determine a final detection result by detecting the sorted node cluster subset. The sorting concept of the node cluster subset may refer to the sorting concept of the node cluster set, which is not further described herein.

In view of the fact that the concept of active optimization (such as manual processing) plays a certain auxiliary role in the precision of keyword selection, the global risk detection description of completion optimization of active optimization for the basic global risk detection description of at least a part of node aggregation sets in a plurality of node clustering sets can be obtained first, and then the local risk detection description of completion optimization for each group of cloud game player behavior data can be obtained based on the determined global risk detection description of completion optimization and the basic global risk detection description of the rest node aggregation sets without completion of active optimization.

The embodiment of the invention can be combined with a big data tag setting thought to perform active optimization, in other words, the basic global risk detection description of each node aggregation set can be input into a relevant big data tag processing model (such as a mature model obtained through training of different types of neural networks), so that active optimization is realized.

For example, it may be understood that, on the basis of determining that a target node cluster set with different behavior categories exists in a plurality of node clusters with an association relationship, the target node cluster set may be designated, and on the basis of acquiring an abnormality for the target node cluster set, active optimization is performed to obtain correct risk detection content. The probability of the attribute corresponding to the optimized node cluster set is set to 1, and the probability of the remaining attributes is set to 0.

It will be appreciated that, based on the proactive optimization of risk detection content, risk detection descriptions for node clusters and behavior feature nodes within a node cluster may also be adapted.

For the embodiment of the invention, the global risk detection description of the node aggregation set can be optimized based on the active optimization thought, and the global risk detection description of the node aggregation set can be optimized according to the risk detection optimization model set in advance, so that the determined local risk detection description for completing optimization of the cloud game player behavior data is more comprehensive.

For embodiments of the present invention, the optimization process described for global risk detection may illustratively include what is recorded in steps 1041-1043.

Step 1041, combining the basic global risk detection description of each node cluster set to obtain global prediction condition offset corresponding to each node cluster set in a risk detection dimension; and obtaining the quantization difference offset corresponding to each node cluster set in the behavior trend dimension according to the quantization difference data between any two node cluster sets which have association relations and are not matched with the same global risk detection description.

For example, the global prediction case offset may be understood as a prediction case error term/prediction case deviation corresponding to each node cluster in the risk detection dimension. The quantitative difference data can be understood as a comparison result between two node aggregation sets which are not matched with the same global risk detection description according to arbitrary and incidence relation. Quantization difference offset can be understood as a contrast error. In addition, the risk detection dimension may be understood as a behavior trend dimension under the risk detection level, and the behavior trend dimension may be understood as a behavior trend dimension under the behavior trend level, so that those skilled in the art may adaptively understand the rest of the content, which is not described in detail in the embodiments of the present invention.

Step 1042, determining a risk detection optimization model (e.g., which can be understood as an energy function) based on the global prediction situation offset and the quantization difference offset.

Step 1043, determining an adjusted global risk detection description determined for each node cluster set on the basis that the output variable of the risk detection optimization model meets the set detection condition.

For example, the output variable of the risk detection optimization model satisfying the set detection condition may be understood as the output variable or the output value of the risk detection optimization model being minimum.

It can be appreciated that for global prediction case migration, not only the basic global risk detection description of the node cluster set, but also the active post-optimization risk detection content, such as: the maximum classification index may be added to the risk detection content of the proactive optimization.

An adjusted global risk detection description for each node cluster set may be obtained by performing a convergence process on the risk detection optimization model (e.g., as may be understood as a minimized risk detection optimization model).

For embodiments of the present invention, optimization of the local risk detection description for cloud game player behavior data may be achieved in conjunction with the adjusted global risk detection description for each node cluster.

It may be understood that, first, based on the correspondence between the first continuous space corresponding to the player behavior feature library and the second continuous space corresponding to the cloud player behavior data, cloud player behavior data corresponding to each behavior feature node covered by the node cluster may be determined from n sets of cloud player behavior data for each, and then, for each set of cloud player behavior data, the adjusted global risk detection description of the node cluster corresponding to the behavior feature node corresponding to the set of cloud player behavior data is combined with the adjusted global risk detection description of the node cluster corresponding to the set of cloud player behavior data, to obtain a local risk detection description optimized for the completion of the set of cloud player behavior data.

Under some independently implementable design considerations, after obtaining a local risk detection description optimized for completion of the cloud player behavior data for each set of cloud players, the method may further include: determining game data protection strategies corresponding to each group of the optimized local risk detection descriptions; activating the game data protection strategy based on the cloud game player behavior data pointed by each group of the optimized local risk detection description.

For the embodiment of the invention, after different game data protection strategies are determined, the game data protection strategy activation can be performed in combination with the behavior habit of cloud game player behavior data, for example, the game data protection strategy is account protection, then the game data protection strategy activation can be performed according to the corresponding non-active period of the cloud game player behavior data, after the game data protection strategy is activated, if the player account corresponding to the cloud game player behavior data is logged in the non-active period, random verification or login prohibition requiring game painter participation can be performed, and thus the targeted and personalized protection of different game data can be realized.

For the embodiment of the present invention, determining the game data protection policy corresponding to each group of the optimized local risk detection descriptions may include the following technical solutions: determining a first attack tendency characteristic according to each group of the optimized local risk detection descriptions, wherein the first attack tendency characteristic is obtained by loading each group of the optimized local risk detection descriptions into a first game data attack analysis network; loading the first attack tendency features and the prior attack tendency feature statistical description into a second game data attack analysis network, and determining second attack tendency features derived by the second game data attack analysis network, wherein the prior attack tendency feature statistical description reflects the relative connection relation of the second attack tendency features derived by the second game data attack analysis network in a prior manner; determining target attack tendency characteristics according to the second attack tendency characteristics; and determining game data protection strategies matched with the attack tendency characteristics from a preset strategy library.

For an embodiment of the present invention, the determining the target attack propensity feature according to the second attack propensity feature includes: determining a plurality of prior second attack propensity characteristics, wherein the prior second attack propensity characteristics are obtained by the second game data attack analysis network according to loaded prior first attack propensity characteristics, and the prior first attack propensity characteristics are obtained by the first game data attack analysis network according to each group of the optimized local risk detection descriptions loaded before the input of each group of the optimized local risk detection descriptions; loading the second attack propensity feature and the plurality of prior second attack propensity features into a third game data attack analysis network, determining third attack propensity features derived by the third game data attack analysis network; and taking the third attack tendency characteristic as a target attack tendency characteristic.

It may be appreciated that, through the above technical solution, a first attack tendency feature is determined, where the first attack tendency feature is an attack tendency feature obtained by loading each set of the optimized local risk detection descriptions into a first game data attack analysis network, then loading the first attack tendency feature and a priori attack tendency feature statistical description into a second game data attack analysis network, determining a second attack tendency feature derived by the second game data attack analysis network, and further determining a target attack tendency feature according to the second attack tendency feature. In this way, for the first attack tendency feature derived through the first game data attack analysis network, the first attack tendency feature is further loaded to the second game data attack analysis network together with the prior attack tendency feature statistical description reflecting the relative connection relation of the second attack tendency feature derived by the second game data attack analysis network, and then the target attack tendency feature is determined according to the second attack tendency feature derived by the second game data attack analysis network, so that the accuracy and the reliability of the target attack tendency feature are ensured, and further the accuracy and the reliability of the game data protection strategy obtained by matching are ensured.

Based on the same inventive concept, fig. 2 shows a block diagram of a cloud game data security device applied to artificial intelligence according to an embodiment of the present invention, where the cloud game data security device applied to artificial intelligence may include the following modules for implementing the relevant method steps shown in fig. 1.

The behavior data determining module 210 is configured to determine a player behavior feature library generated according to n sets of cloud game player behavior data and a base local risk detection description for each set of cloud game player behavior data.

The feature node arrangement module 220 is configured to arrange each behavior feature node covered by the player behavior feature library into a plurality of node aggregation sets.

The cluster set detection module 230 is configured to perform global risk detection on each node cluster set obtained by the overall process based on the basic local risk detection description, so as to obtain a basic global risk detection description of each node cluster set.

The detection description obtaining module 240 is configured to obtain, in combination with the basic global risk detection description of each node cluster set, a local risk detection description optimized for each cloud game player behavior data.

The related embodiments applied to the present invention can achieve the following technical effects: the embodiment of the invention optimizes the basic local risk detection description of the cloud game player behavior data under the local feature environment based on the basic global risk detection description of the node aggregation set under the global feature environment, and the local risk detection description of the cloud game player behavior data of each group of cloud game player behavior data can be normally used in the subsequent game data protection process due to the fact that the global feature environment has a plurality of behavior feature nodes corresponding to one node aggregation set and the cloud game player behavior data corresponding to each behavior feature node can be subjected to packaged risk detection processing through the technical thought, so that the precision and the credibility of the optimized local risk detection description are guaranteed, and in addition, the risk detection of the cloud game player behavior data can be effectively realized through the risk detection processing thought, and the normal use of the cloud game player behavior data of each group of cloud game player behavior data can be guaranteed.

The foregoing is merely a specific embodiment of the present application. Variations and alternatives will occur to those skilled in the art from the detailed description provided herein and are intended to be included within the scope of the present application.

Claims (10)

1. A cloud game data security protection method applied to artificial intelligence, which is characterized by at least comprising the following steps:

determining a player behavior feature library generated from n sets of cloud player behavior data, and a base local risk detection description for each set of cloud player behavior data; wherein n is a positive integer;

sorting each behavior feature node covered by the player behavior feature library into a plurality of node aggregation sets; carrying out global risk detection on each node cluster obtained through the overall management through the basic local risk detection description to obtain basic global risk detection description of each node cluster;

and acquiring a local risk detection description optimized for the completion of the cloud game player behavior data of each group by combining the basic global risk detection description of each node aggregation set.

2. The method of claim 1, wherein said obtaining a locally risk detection description optimized for completion of said each cloud player behavior data in conjunction with said base global risk detection description for each of said node clusters comprises:

Determining a global risk detection description for completing optimization; the global risk detection description after optimization is active global risk detection description for basic global risk detection description of at least a part of node cluster sets in the plurality of node cluster sets;

and obtaining the optimized local risk detection description aiming at the behavior data of each cloud game player through the optimized global risk detection description and the basic global risk detection description of the residual node aggregation set which does not finish the active optimization.

3. The method of claim 1, wherein said obtaining a locally risk detection description optimized for completion of said each cloud player behavior data in conjunction with said base global risk detection description for each of said node clusters comprises:

combining the basic global risk detection description of each node cluster set to obtain global prediction condition offset corresponding to each node cluster set in the risk detection dimension;

obtaining quantized difference offset corresponding to each node cluster set in the behavior trend dimension according to quantized difference data between two node cluster sets which are random and have association relations and are not matched with the same global risk detection description;

Determining a risk detection optimization model through the global prediction situation offset and the quantitative difference offset;

determining an adjusted global risk detection description determined for each node cluster on the basis of ensuring that the output variable of the risk detection optimization model meets a set detection condition;

and acquiring a local risk detection description optimized for the completion of the cloud game player behavior data of each group by combining the adjusted global risk detection description for each node aggregation set.

4. The method of claim 1, wherein said obtaining a locally risk detection description optimized for completion of said each cloud player behavior data in conjunction with said base global risk detection description for each of said node clusters comprises:

for each node aggregation set, determining cloud game player behavior data corresponding to each behavior feature node covered by the node aggregation set from the n groups of cloud game player behavior data based on the corresponding situation between a first continuous space corresponding to the player behavior feature library and a second continuous space corresponding to the cloud game player behavior data;

aiming at the behavior data of each cloud game player, combining the basic global risk detection description of the node aggregation set corresponding to the behavior feature nodes corresponding to the behavior data of the cloud game player to acquire the local risk detection description for completing optimization of the behavior data of the cloud game player;

Wherein the basic local risk detection is described as a result of risk detection of the cloud player behavior data of each set by means of a basic game risk detection model, and the obtaining of the local risk detection description optimized for the completion of the cloud player behavior data of each set further comprises: and configuring the basic game risk detection model through the optimized local risk detection description of each group of cloud game player behavior data to obtain a configured game risk detection model.

5. The method of claim 1, wherein the sorting each behavior feature node covered by the player behavior feature library into a plurality of node clusters comprises:

randomly determining target number of behavior feature nodes from each behavior feature node covered by the player behavior feature library, and taking the key feature information of each determined behavior feature node as the reference feature information of the node aggregation set to be clustered and arranged;

determining the key feature information of each behavior feature node in each behavior feature node, sequentially comparing the key feature information with each reference feature information in a quantitative comparison mode, and sorting the behavior feature nodes into a node aggregation set to which the corresponding reference feature information with the quantitative comparison result meeting the specified requirements belongs;

Obtaining current reference characteristic information corresponding to each node aggregation set, and combining the current reference characteristic information, and jumping to the step of arranging the node aggregation sets until the node aggregation sets accord with the clustering evaluation index;

the obtaining the current reference characteristic information corresponding to each node aggregation set comprises the following steps: and combining key characteristic information of each behavior characteristic node covered by each node cluster set to obtain depolarization characteristic information of each behavior characteristic node, and taking the depolarization characteristic information as current reference characteristic information corresponding to the node cluster set.

6. The method of claim 5, wherein the key feature information comprises a key indicative feature and a target differential description, and the reference feature information comprises a reference indicative feature and a reference differential description; the determining the quantitative comparison result between the key feature information of each behavior feature node in each behavior feature node and each reference feature information in turn comprises the following steps:

determining a first quantitative comparison result between the behavior feature node and a node cluster to which each piece of reference feature information belongs by combining key indication type features of the behavior feature node and reference indication type features of each piece of reference feature information aiming at each behavior feature node in the behavior feature nodes, and determining a second quantitative comparison result between the behavior feature node and the node cluster to which each piece of reference feature information belongs by combining target differential description of the behavior feature node and reference differential description of each piece of reference feature information;

And determining the key characteristic information of the behavior characteristic node through the first quantitative comparison result and the importance index matched with the first quantitative comparison result and the second quantitative comparison result and the importance index matched with the second quantitative comparison result, and the quantitative comparison result between the key characteristic information and each piece of reference characteristic information in sequence.

7. The method according to claim 1, wherein the performing global risk detection on each node cluster obtained by the ensemble through the basic local risk detection description to obtain a basic global risk detection description of each node cluster includes:

for each behavior feature node, determining a mapping data set corresponding to the behavior feature node in n groups of cloud game player behavior data based on the corresponding situation between a first continuous space corresponding to the player behavior feature library and a second continuous space corresponding to the cloud game player behavior data;

determining a basic global risk detection description of the behavior feature node in combination with the basic local risk detection description of the mapping data set corresponding to the behavior feature node in the n groups of cloud game player behavior data;

aiming at each node aggregation set obtained by arrangement, combining the basic global risk detection description of each behavior feature node covered by the node aggregation set, and determining the basic global risk detection description of the node aggregation set;

Wherein said determining, in combination with said base local risk detection description of said mapped dataset corresponding to said behavior feature node in said n sets of cloud game player behavior data, a base global risk detection description of said behavior feature node comprises: for each behavior feature node, determining keyword statistical results corresponding to various basic local risk detection descriptions by combining basic local risk detection descriptions of corresponding mapping data sets of the behavior feature node in the n groups of cloud game player behavior data; and combining keyword statistical results corresponding to various basic local risk detection descriptions to determine basic global risk detection descriptions of the behavior feature nodes.

8. The method of claim 7, wherein determining the basic global risk detection description of the behavior feature node in combination with the keyword statistics of the corresponding types of basic local risk detection descriptions comprises: determining a basic local risk detection description of which the keyword statistical result meets a set statistical condition as a basic global risk detection description of the behavior feature node;

the determining, for each node cluster obtained by the sorting, a basic global risk detection description of the node cluster in combination with a basic global risk detection description of each behavior feature node covered by the node cluster includes: combining basic global risk detection descriptions of each behavior feature node covered by each node aggregation set, and determining statistical values of the behavior feature nodes corresponding to various basic global risk detection descriptions; and determining the basic global risk detection description of the node aggregation set according to the basic global risk detection description of the class with the maximum statistical value of the corresponding behavior feature nodes.

9. The method of claim 7, wherein determining the basic global risk detection description of the behavior feature node in combination with the keyword statistics of the corresponding types of basic local risk detection descriptions comprises: carrying out quantitative analysis processing on keyword statistical results and keyword summation results corresponding to various basic local risk detection descriptions, and determining that the behavior feature nodes are matched with quantitative prediction results of each type of basic local risk detection description; determining basic global risk detection description of the behavior feature node according to quantitative prediction results matched with various basic local risk detection descriptions;

the determining, for each node cluster obtained by the sorting, a basic global risk detection description of the node cluster in combination with a basic global risk detection description of each behavior feature node covered by the node cluster includes: for each node cluster set obtained by arrangement, combining a quantized prediction result which corresponds to each behavior feature node covered by the node cluster set and is matched with each type of basic local risk detection description, and an importance index which is previously set for the behavior feature node and corresponds to each type of basic local risk detection description, and determining the quantized prediction result which corresponds to each type of basic local risk detection description; combining the quantized prediction results of the node cluster set matched with each type of basic local risk detection description to determine basic global risk detection description of the node cluster set;

Wherein, on the basis of determining that the quantized predicted outcome of any node cluster matched with the multi-class basis local risk detection description exceeds the specified predicted outcome, the method further comprises: for any node cluster set, the node cluster set is arranged into a plurality of node cluster subsets;

the global risk detection is performed on each node cluster obtained through the basic local risk detection description, and basic global risk detection description of each node cluster is obtained, including: carrying out global risk detection on each node cluster subset obtained through the overall management through the basic local risk detection description to obtain basic global risk detection description of each node cluster subset;

the obtaining, in combination with the global risk detection description of each node aggregation set, a local risk detection description optimized for completion of the cloud game player behavior data of each group includes: and obtaining the optimized local risk detection description aiming at the behavior data of each cloud game player by the basic global risk detection description of each node cluster subset.

10. An artificial intelligence server, comprising: a memory and a processor; the memory is coupled to the processor; the memory is used for storing computer program codes, and the computer program codes comprise computer instructions; wherein the computer instructions, when executed by the processor, cause the artificial intelligence server to perform the method of any one of claims 1-9.