Capturing ordinal historical dependence in graphical event models with tree representations

1 . A computer system comprising:

a processor; and

a memory in communication with the processor, the memory containing program instructions that, when executed by the processor, are configured to cause the processor to perform a method of learning a tree ordinal graphical event model from temporal event data comprising a time-stamped event stream, the method comprising:

modeling temporal relationships between one or more events in the temporal event data, wherein each event is represented by a node;

learning an ordered graphical event model (OGEM) graph, including capturing ordinal historical dependence of the one or more events for a time window prior to each occurrence of the event, wherein the OGEM graph is a structure configured to illustrate influence of history of occurrences on rates of arrival of event types;

automatically learning a parameter sharing architecture, including leveraging the OGEM graph and learning order dependent statistical and causal co-occurrence relationships among event types;

computing, based on a tracked history of event occurrences in real time, a control action for a corresponding predictive event type;

generating, selectively, a control signal based on the computed control action;

applying the generated control signal as a feedback signal to selectively control a corresponding event injection; and

dynamically issuing the control signal to an operatively coupled event device associated with at least one event type reflected in the learned parameter sharing architecture, wherein the control signal is configured to selectively control the event injection in accordance with the captured ordinal historical dependence, and wherein the event injection is configured to manage event behavior within the temporal event data,

wherein the learned parameter sharing architecture is configured to capture event dynamics by one or more conditional intensity functions to measure a rate of occurrence of an event label.

2 . The computer system of claim 1 , wherein the learned parameter sharing architecture is a tree based OGEM, and further comprises assigning each leaf in the tree with a label from an event label set, wherein the event label set is restricted by a subset of labels or a wildcard to accept any label in the event label set, and wherein the leaves of the tree represent an order sensitive parameterization for events of the OGEM.

3 . The computer system of claim 1 , wherein managing the event behavior comprises an action selected from the group consisting of modifying, delaying, or mitigating effects of one event on another event.

4 . The computer system of claim 1 , wherein the temporal event data comprises asynchronous time-stamped events with irregular inter-event spacing.

5 . The computer system of claim 1 , wherein learning the parameter sharing architecture includes applying a masking function to derive an active order of event types.

6 . A computer program product comprising:

a computer readable storage device; and

program code embodied with the computer readable storage device, the program code executable by a processor to:

model temporal relationships between one or more events in received temporal event data, wherein each event is represented by a node;

learn an ordered graphical event model (OGEM) graph, including capture ordinal historical dependence of the one or more events for a time window prior to each occurrence of the event, wherein the OGEM graph is a structure configured to illustrate influence of history of occurrences on rates of arrival of event types;

automatically learn a parameter sharing architecture, including leverage the OGEM graph and learning order dependent statistical and causal co-occurrence relationships among event types;

compute, based on a tracked history of event occurrences in real time, a control action for a corresponding predictive event type;

generate, selectively, a control signal based on the computed control action;

apply the generated control signal as a feedback signal to selectively control a corresponding event injection;

dynamically issue the control signal to an operatively coupled event device associated with at least one event type reflected in the learned parameter sharing architecture, wherein the control signal configured to selectively control the event injection in accordance with the captured ordinal historical dependence, and wherein the event injection is configured to manage event behavior within the temporal event data,

wherein the learned parameter sharing architecture is configured to capture event dynamics by one or more conditional intensity functions to measure a rate of occurrence of an event label.

7 . The computer program product of claim 6 , wherein the learned parameter sharing architecture is a tree based OGEM, and further comprising program code configured to assign each leaf in the tree with a label from an event label set, wherein the event label set is restricted by a subset of labels or a wildcard to accept any label in the event label set, and wherein the leaves of the tree represent an order sensitive parameterization for events of the OGEM.

8 . A method comprising:

modeling temporal relationships between one or more events in received temporal event data, wherein each event is represented by a node;

learning an ordered graphical event model (OGEM) graph, including capturing ordinal historical dependence of the one or more events for a time window prior to each occurrence of the event, wherein the OGEM graph is a structure configured to illustrate influence of history of occurrences on rates of arrival of event types;

automatically learning a parameter sharing architecture, including leveraging the OGEM graph and learning order dependent statistical and causal co-occurrence relationships among event types;

computing, based on a tracked history of event occurrences in real time, a control action for a corresponding predictive event type;

generating, selectively, a control signal based on the computed control action;

applying the generated control signal as a feedback signal to selectively control a corresponding event injection;

dynamically issuing the control signal to an operatively coupled event device associated with at least one event type reflected in the learned parameter sharing architecture, and based on the history of occurrence rates, wherein the control signal configured to selectively control the event injection in accordance with the captured ordinal historical dependence, and wherein the event injection is configured to manage event behavior within the temporal event data,

wherein the learned parameter sharing architecture is configured to capture event dynamics by one or more conditional intensity functions for measuring a rate of occurrence of an event label.

9 . The method of claim 8 , wherein the learned parameter sharing architecture is a tree based OGEM, and further comprising assigning each leaf in the tree with a label from an event label set, wherein the event label set is restricted by a subset of labels or a wildcard to accept any label in the event label set, and wherein the leaves of the tree represent an order sensitive parameterization for events of the OGEM.

10 . The method of claim 8 , wherein managing the event behavior comprises an action selected from the group consisting of modifying, delaying, or mitigating effects of one event on another event.

11 . The method of claim 8 , wherein the temporal event data comprises asynchronous time-stamped events with irregular inter-event spacing.

12 . The method of claim 8 , wherein learning the parameter sharing architecture includes applying a masking function to derive an active order of event types.