Dynamic rule allocation for visitor identification

1. A non-transitory computer storage media storing computer-useable instructions that, when used by one or more computing devices, cause the one or more computing devices to perform operations comprising:

computing, at a server device, effectives scores for each cookie-stitching rule within a set of cookie-stitching rules by applying each cookie-stitching rule to each cookie with a data set, wherein each effectiveness score comprises a binary-classification evaluation metric;

computing, at the server device, a cookie representation for a received cookie under a numeric vector space, the cookie representation being computed using cookie statistics including a frequency of hits in log data received from a web server device;

selecting, at the server device, a particular cookie-stitching rule from the set of cookie-stitching rules to apply to the received cookie, wherein the particular cookie-stitching rule is selected based on having a highest predicted effectiveness score for identifying cookies with cookie representations similar to the cookie representation relative to effectiveness scores of other cookie-stitching rules from the set of cookie-stitching rules, wherein the highest predicted effectiveness score comprises a highest binary-classification evaluation metric; and

applying, at the server device, the particular cookie-stitching rule to the cookie representation to identify a website visitor associated with the received cookie.

2. The non-transitory computer storage media of claim 1 , wherein the particular cookie-stitching rule is further selected based on a selected goal.

3. The non-transitory computer storage media of claim 2 , wherein the selected goal is one of precision, recall, or F-beta measure, wherein:

when the selected goal is precision, the selected rule has the highest precision;

when the selected goal is recall, the selected rule has the highest recall; and

when the selected goal is F-beta measure, the selected rule has the highest F-beta measure.

4. The non-transitory computer storage media of claim 1 , wherein the log data may include zip codes, IP addresses, or user agents.

5. The computerized method of claim 1 , further comprising adding each binary-classification evaluation metric to the data set.

6. A computerized method for identifying a website visitor, the computerized method comprising:

receiving, via a first computing process performed at a server device, a set of log data for a website from a web server device, the set of log data including cookies and an identification of a website visitor corresponding to each cookie;

computing, via a second computing process performed at the server device, a cookie representation for each cookie under a numeric vector space, the cookie representation being computed using cookie statistics including a frequency of hits in the log data;

evaluating, via a third computing process preformed at the server device, cookie-stitching rules from a set of cookie-stitching rules by applying each cookie-stitching rule from the set of cookie-stitching rules to the cookie representations to determine an effectiveness score for each cookie-stitching rule;

training, via a fourth computing process performed at the server device, a classifier of effectiveness scores for each cookie-stitching rule of the set of cookie-stitching rules, each effectiveness score being a measurement of the effectiveness of each cookie-stitching rule at identifying website visitors based on the cookie representation and comprising a binary-classification evaluation metric; and

utilizing, via a fifth computing process performed at the server device, the classifier to select a most effective cookie-stitching rule for log data corresponding to an unknown website visitor, the most effective cookie-stitching rule being a cookie-stitching rule with the highest cookie-stitching effectiveness score relative to other cookie-stitching rules from the set of cookie-stitching rules, wherein the highest effectiveness score comprises a highest binary-classification evaluation metric.

7. The computerized method of claim 6 , further comprising building for each cookie-stitching rule a regression tree.

8. The computerized method of claim 7 , further comprising applying the most effective cookie-stitching rule to a cookie representation to identify the unknown website visitor.

9. The computerized method of claim 6 , wherein the classifier is trained for different goals including recall, precision, or F-beta measure.

10. The computerized method of claim 6 , further comprising adding each binary-classification evaluation metric to a data set.

11. A computerized system comprising:

a processor; and

a non-transitory computer storage media storing computer-useable instructions that, when used by the processor, cause the processor to:

predict effectiveness scores for each cookie-stitching rule in a set of cookie-stitching rules for a cookie representation, the cookie representation being computed using cookie statistics including a frequency of hits in log data, the effectiveness scores being predicted by applying each cookie-stitching rule to a cookie representation, wherein each effectiveness score measuring measures an effectiveness for identifying cookies with cookie representations similar to the cookie representation and comprises a binary-classification evaluation metric;

select, from the set of cookie-stitching rules, the cookie-stitching rule with a highest effectiveness score, the highest effectiveness score comprising a highest binary-classification evaluation metric and indicating that the cookie-stitching rule is most effective, relative to other cookie-stitching rules within the set, at identifying website visitors associated with cookies with similar cookie representations to the cookie representation; and

apply the cookie-stitching rule to a received cookie to identify an unknown website visitor associated with the received cookie.

12. The computerized system of claim 11 , wherein the computer-usable instructions further cause the processor to add each binary-classification evaluation metric to a data set.

13. The computerized system of claim 11 , wherein the effectiveness scores vary based on a selected goal, the selected goal including one of recall, precision, or F-beta measure.

14. The computerized system of claim 11 , wherein the computer-useable instructions further cause the processor to train a predictor of effectiveness scores, the predictor trained utilizing a set of data from known website visitors.