Automated detection of emergent behaviors in interactive agents of an interactive environment
Various aspects of the subject technology relate to systems, methods, and machine-readable media for automated detection of emergent behaviors in interactive agents of an interactive environment. The disclosed system represents an artificial intelligence based entity that utilizes a trained machine-learning-based clustering algorithm to group users together based on similarities in behavior. The clusters are processed based on a determination of the type of activity of the clustered users. In order to identify new categories of behavior and to label those new categories, a separate cluster analysis is performed using interaction data obtained at a subsequent time. The additional cluster analysis determines any change in behavior between the clustered sets and/or change in the number of users in a cluster. The identification of emergent user behavior enables the subject system to treat users differently based on their in-game behavior and to adapt in near real-time to changes in their behavior.
1. A computer-implemented method, comprising:
obtaining first interaction data of a participant device in a multiuser session of an interactive environment based on interactions between an interactive agent associated with a user and the interactive environment;
obtaining second interaction data of the participant device based on subsequent interactions between the interactive agent and the interactive environment;
analyzing the first interaction data and the second interaction data with a machine-learning-based cluster analysis model comprising an output of an artificial neural network;
determining a behavior signature of each of a plurality of interactive agents based on the analyzed first and second interaction data;
detecting an emergent behavior associated with a cluster comprised of a subset of interactive agents of the plurality of interactive agents with similar behavior signatures, the detecting comprising:
comparing a first metric value of the interactive agent from the first interaction data to a second metric value of the interactive agent from the second interaction data;
determining a difference between the first metric value and the second metric value, wherein the first metric value corresponds to a first cluster of the plurality of clusters and the second metric value corresponds to a second cluster of the plurality of clusters, the second cluster being different from the first cluster; and
determining, from the first and second interaction data corresponding to a same cluster, that a behavior signature of the interactive agent corresponds to the emergent behavior based on the determined difference and metric values of other interactive agents, the other interactive agents having similar attributes as that of the interactive agent; and
adjusting a status of a user corresponding to an interactive agent associated with the detected emergent behavior.
2. The computer-implemented method of claim 1 , wherein adjusting the status of the user comprises:
removing the interactive agent associated with the user from the multiuser session when the interactive agent is associated with the detected emergent behavior.
3. The computer-implemented method of claim 1 , further comprising:
modifying a state of the interactive environment based on an occurrence of the emergent behavior, wherein the modified state of the interactive environment indicates a change in types of interactions between the interactive agent and the interactive environment that are allowed during the multiuser session.
4. The computer-implemented method of claim 1 , wherein obtaining the interaction data comprises measuring a number of interactions between the interactive agent and the interactive environment, and wherein each of the interactions may correspond to different types of activity of the interactive environment.
5. The computer-implemented method of claim 1 , wherein analyzing the first interaction data and the second interaction data comprises:
applying a nearest neighbor clustering algorithm with the machine-learning-based cluster analysis model; and
determining one or more clusters that correspond to different types of behaviors from the analyzed first and second interaction data.
6. The computer-implemented method of claim 1 , wherein analyzing the first interaction data and the second interaction data comprises:
processing the first interaction data and the second interaction data with the machine-learning-based cluster analysis model to predict similarities and differences between interactions of different interactive agents for each of a plurality of metrics being measured.
7. The computer-implemented method of claim 6 , wherein analyzing the first interaction data and the second interaction data comprises:
determining a plurality of items from the first interaction data and the second interaction data, wherein each of the plurality of items represents a different interactive agent of the interactive environment;
mapping the plurality of items to an N-dimensional space, where N is a positive integer; and
setting a metric value for N different dimensions of the N-dimensional space for each item of the plurality of items, wherein each of the plurality of metrics corresponds to one different dimension of the N-dimensional space.
8. The computer-implemented method of claim 7 , wherein analyzing the first interaction data and the second interaction data comprises:
calculating a proximity between data points on the N-dimensional space to group items of the plurality of items into a plurality of clusters, wherein each of the plurality of clusters has a center with an n-dimensional location;
calculating a distance between centers of the plurality of clusters for each of the plurality of clusters;
determining a closest neighboring cluster of the plurality of clusters based on the calculated distance for each of the plurality of clusters;
determining one or more dimensions of the N-dimensional space that have a greatest distance to the closest neighboring cluster; and
classifying a cluster with a label that identifies the cluster based on the determined one or more dimensions.
9. The computer-implemented method of claim 1 , wherein detecting the emergent behavior comprises:
comparing a first plurality of clusters mapped from the first interaction data to a second plurality of clusters mapped from the second interaction data; and
determining at least one of the second plurality of clusters that is not present in the first plurality of clusters from the comparison, wherein the at least one of the second plurality of clusters represents a group of interactive agents that have a pattern of behavior that is not previously present in the first plurality of clusters, and wherein the pattern of behavior that is not previously present represents the detected emergent behavior.
10. The computer-implemented method of claim 1 , wherein adjusting the status of the user comprises:
marking a user account with an association to the emergent behavior when the user account corresponds to the interactive agent associated with the emergent behavior; and
purging the user account marked with the association to the emergent behavior; and
purging the interactive agent associated with the emergent behavior from an interactive agent log identifying a plurality of interactive agents instantiated in the interactive environment when the user account marked with the association to the emergent behavior is purged.
11. A non-transitory computer readable storage medium is provided including instructions that, when executed by a processor, cause the processor to perform a method, the method comprising:
obtaining first interaction data of a participant device in a multiuser session of an interactive environment based on interactions between an interactive agent associated with a user and an interactive environment;
obtaining second interaction data of the participant device based on subsequent interactions between the interactive agent and the interactive environment;
analyzing the first interaction data and the second interaction data with a machine-learning-based cluster analysis model comprising an output of an artificial neural network;
determining a behavior signature of each of a plurality of interactive agents based on the analyzed first and second interaction data;
detecting an emergent behavior associated with a cluster comprised of a subset of interactive agents of the plurality of interactive agents with similar behavior signatures the detecting comprising:
comparing a first metric value of the interactive agent from the first interaction data to a second metric value of the interactive agent from the second interaction data;
determining a difference between the first metric value and the second metric value, wherein the first metric value corresponds to a first cluster of the plurality of clusters and the second metric value corresponds to a second cluster of the plurality of clusters, the second cluster being different from the first cluster; and
determining, from the first and second interaction data corresponding to a same cluster, that a behavior signature of the interactive agent corresponds to the emergent behavior based on the determined difference and metric values of other interactive agents, the other interactive agents having similar attributes as that of the interactive agent; and
adjusting a status of a user corresponding to an interactive agent associated with the detected emergent behavior.
12. The non-transitory computer readable storage medium of claim 11 , wherein adjusting the status of the user comprises:
removing the interactive agent associated with the user from the multiuser session when the interactive agent is associated with the detected emergent behavior.
13. The non-transitory computer readable storage medium of claim 11 , wherein the method further comprises:
modifying a state of the interactive environment based on an occurrence of the emergent behavior, wherein the modified state of the interactive environment indicates a change in types of interactions between the interactive agent and the interactive environment that are allowed during the multiuser session.