Back to results
US 9,430,535 B2Granted
Method and apparatus for normalizing and predicting time series data
AssigneeAdobe Systems Incorporated
InventorAnubha Rastogi
Life of this patent
Estimating the term…
Abstract
A computer implemented method and apparatus for normalizing and predicting time series data. The method comprises accessing collected data comprising a plurality of intervals; defining a variation for each interval in the plurality of intervals, wherein each variation is a cost value; clustering the cost values, wherein clustering identifies anomalies in the collected data; correcting the anomalies in the collected data; and creating a set of normalized data from the corrected data.
Claims20
A computer implemented method comprising:
- accessing website visitor data, the website visitor data comprising a plurality of time-series intervals, wherein each of the time-series intervals reflect a period of time that a website is available to receive visitors;
- determining, by at least one processor and using a cost function, a cost value for each time-series interval in the plurality of time-series intervals, wherein the cost function compares website visitor data from at least two time-series intervals;
- organizing, by the at least one processor, the determined cost values into multiple cost value clusters using a clustering algorithm;
- identifying at least one cost value cluster from the multiple cost value clusters by identifying at least one typical cost value cluster based on user input;
- identifying, by the at least one processor, one or more anomalous clusters from the multiple cost value clusters, the one or more anomalous clusters representing at least one anomaly in the website visitor data;
- modifying at least one cost value corresponding to the at least one anomaly in the website visitor data based on the at least one cost value cluster, wherein modifying the at least one cost value comprises:
- generating a mean cost value from the at least one typical cost value cluster, and
- replacing the at least one cost value corresponding to the at least one anomaly with the mean cost value;
- creating, by the at least one processor, a set of normalized data from the at least one modified cost value by applying the cost function to the modified at least one cost value;
- predicting website visitor data for a future time-series interval based on the set of normalized data;
- and
- selecting, by the at least one processor, a digital advertisement for the website using the predicted website visitor data.
9 more elements
A system comprising:
- at least one processor;
- and
- at least one non-transitory computer readable storage medium storing instructions thereon that, when executed by the at least one processor, cause the system to:
- access website visitor data comprising a plurality of time-series intervals, wherein each of the time-series intervals reflect a period of time that a website is available to receive visitors;
- determine, using a cost function, a cost value for each time-series interval in the plurality of time-series intervals, wherein the cost function compares website visitor data from at least two time-series intervals;
- organize the determined cost values into multiple cost value clusters using a clustering algorithm;
- identify at least on cost value cluster from the multiple cost value clusters by identifying at least one typical cost value cluster based on user input;
- identify one or more anomalous clusters from the multiple cost value clusters, the one or more anomalous clusters representing at least one anomaly in the website visitor data;
- modify at least one cost value corresponding to the at least one anomaly in the website visitor data based on the at least one cost value cluster by performing steps comprising:
- generating a mean cost value from the at least one typical cost value cluster;
- and
- replacing the at least one cost value corresponding to the at least one anomaly with the mean cost value;
- create a set of normalized data from the at least one modified cost value by applying the cost function to the modified at least one cost value;
- predict website visitor data for a future time-series interval based on the set of normalized data;
- and
- select a digital advertisement for the website using the predicted website visitor data.
13 more elements
A non-transitory computer readable medium storing instructions thereon that, when executed by at least one processor, cause a computer system to:
- access website visitor data, the website visitor data comprising a plurality of time-series intervals, wherein each of the time-series intervals reflect a period of time that a website is available to receive visitors;
- determine, using a cost function, a cost value for each time-series interval in the plurality of time-series intervals, wherein the cost function compares website visitor data from at least two time-series intervals;
- organize the determined cost values into multiple cost value clusters using a clustering algorithm;
- identifying at least one cost value cluster from the multiple cost value clusters by identifying at least one typical cost value cluster based on user input;
- identify one or more anomalous clusters from the multiple cost value clusters, the one or more anomalous clusters representing at least one anomaly in the website visitor data;
- modify at least one cost value corresponding to the at least one anomaly in the website visitor data based on the at least one cost value cluster by performing steps comprising:
- generating a mean cost value from the at least one typical cost value cluster;
- and
- replacing the at least one cost value corresponding to the at least one anomaly with the mean cost value;
- create a set of normalized data from the at least one modified cost value by applying the cost function to the modified at least one cost value;
- predict website visitor data for a future time-series interval based on the set of normalized data;
- and
- select a digital advertisement for the website using the predicted website visitor.
10 more elements
Dependencies are inferred from the claim text.