| US 7,519,566 B2 | ||
| Method and apparatus for automatically and continuously updating prediction models in real time based on data mining | ||
| Sergey A. Prigogin, Foster City, Calif. (US); and Michel Adar, Palo Alto, Calif. (US) | ||
| Assigned to Oracle International Corporation, Redwood Shores, Calif. (US) | ||
| Filed on Nov. 02, 2004, as Appl. No. 10/980,421. | ||
| Claims priority of provisional application 60/544192, filed on Feb. 11, 2004. | ||
| Prior Publication US 2005/0177414 A1, Aug. 11, 2005 | ||
| Int. Cl. G06E 1/00 (2006.01) | ||
| U.S. Cl. 706—21 [707/100] | 43 Claims |
| 1. A computer-implemented method of updating at least one prediction model for use by at least one interactive server, wherein
each interactive server performs a plurality of actions in the context of a plurality of input attribute values of an input
dataset and wherein the actions are selected based on each prediction model, the method comprising:
(a) automatically and continually obtaining contextual data from the interactive server as it performs the plurality of actions,
wherein the contextual data indicates at least which action was performed, which input attribute values are present for each
action that was performed, and which outcome is achieved for each action was performed;
(b) automatically and continually updating a learning model based on all of the obtained contextual data, wherein the learning
model predicts a probability of each of a plurality of specific outcomes occurring for each of a plurality of specific actions
being performed by the interactive server when specific combinations of one or more input attribute values are present, wherein
the updating of the learning model is based on counts of each attribute value as it is present along with each of the other
attribute values;
(c) generating an updated prediction model, wherein generating comprises,
determining a correlation between one or more values of input attributes and a target of prediction,
comparing the correlation with a threshold,
eliminating each value of input attributes from the input dataset if the value of input attribute falls to satisfy the threshold,
and
in response to eliminating, generating the updated prediction model based on the input dataset;
(d) generating a prediction of a probability of an outcome using the updated prediction model;
(e) selecting an action of the plurality of actions based on the prediction; and
(f) performing the action,
wherein operations (b) and (c) are performed each time a condition is met, the condition being selected from a group consisting
of (i) a predetermined level of contextual data has been obtained, (ii) a predetermined number of actions have been performed,
(iii) a predetermined time period has expired, and (iv) a number of new input attributes from the collected contextual data
has reached a predetermined percentage of a total number of the input attributes or a predetermined minimum number of new
input attributes has been reached.
|