Surfaid predictor: web-based system for predicting surfer behavior

• US 6,338,066 B1
• Filed: 09/25/1998
• Issued: 01/08/2002
• Est. Priority Date: 09/25/1998
• Status: Expired due to Term

• Current Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Original Assignee: International Business Machines Corporation
• Inventors: Martin, David Charles, Wolpert, David Hilton, Miranda, Hansel Joseph, Plutowski, Mark Paul, Wheeler, Kevin, Spangler, William Scott, Vaithyanathan, Shivakumar
• Primary Examiner(s): HOMERE, JEAN RAYMOND
• Application Number:
  US09/160,828
• Time in Patent Office: 1,201 Days
• Field of Search: 707/5, 707/6, 707/10, 707/102, 707/201, 709/203, 709/217, 709/223
• US Class Current: 707/708
• CPC Class Codes: G06F 16/958 Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access
• CPC Class Codes: Y10S 707/99936 Pattern matching access
• CPC Class Codes: Y10S 707/99943 Generating database or data structure, e.g. via user interface

Abstract

Given a log of previous web-surfer behavior listing the order in which each surfer downloaded specific items at the web site, and given a meaningful classification of those same items, future surfer behavior is predicted by the present invention. The algorithm utilizes a quantitative model relating items downloaded prior to some specified event to items downloaded after that same event. When the model is applied to a new surfer'"'"'s session prior to an analogous event, the present invention predicts the likely behavior of the surfer subsequent to that event. The predicted behavior is then further analyzed to derive a quantitative value for the utility of the expected behavior.

By randomly selecting sample sessions from a web log, multiple models of surfer behavior can be generated. The multiple models can then be applied to a new surfer'"'"'s session to produce a predicted behavior/utility distribution and thus a confidence interval for the predicted behavior/utility.

Claims

1. A method for dynamically generating views on web pages based on a log of user activities and site content information at a website, comprising:

recording a log of users'"'"' activities at a website, wherein the log is sessionalized for each user;

classifying a plurality of web site items from each session into item groups;

categorizing the item groups into at least three classifications, the classifications comprising features, actions, and behaviors;

grouping the features and behaviors by the actions associated with the features and behaviors;

producing a feature vector for each action occurrence in each session to describe a users behavior up to a point each action occurred;

producing a behavioral vector for each action occurrence in each session to describe which item groups a predictive model will predict the users will select subsequent to each action; and

training the predictive model to predict the user'"'"'s behavior and generate a behavior vector for a given a feature vector.

2. The method of claim 1, further comprising random sampling the features, actions, and behaviors.

3. The method of claim 2, wherein the grouping sorts the sampled features and behaviors by the associated action.

4. The method of claim 1, wherein the models are linear.

5. The method of claim 1, wherein the training the predictive model further comprises:

defining a set of actions that can be downloaded; and

multiplying the feature vector and the set of actions to create a behavior vector.

6. An apparatus for predicting behavior at a website, comprising:

a network server with a memory; and

a computer program, performed by the network server, for recording a log of users'"'"' activities at a website, wherein the log is sessionalized for each user;

classifying a plurality of web site items from each session into item groups;

categorizing the item groups into at least three classifications, the classifications comprising features, actions, and behaviors;

grouping the features and behaviors by the actions associated with the features and behaviors;

producing a feature vector for each action occurrence in each session to describe a user'"'"'s behavior up to a point each action occurred;

producing a behavioral vector for each action occurrence in each session to describe which item groups a predictive model will predict the users will select subsequent to each action; and

training the predictive model to predict the user'"'"'s behavior and generate a behavior vector for a given a feature vector.

7. An article of manufacture comprising a computer program carrier readable by a computer and embodying one or more instructions executable by the computer to perform method steps for configuring an operating system, the method comprising:

recording a log of users'"'"' activities at a website, wherein the log is sessionalized for each user;

classifying a plurality of web site items from each session into item groups;

categorizing the item groups into at least three classifications, the classifications comprising features, actions, and behaviors;

grouping the features and behaviors by the actions associated with the features and behaviors;

producing a feature vector for each action occurrence in each session to describe a users behavior up to a point each action occurred;

producing a behavioral vector for each action occurrence in each session to describe which item groups a predictive model will predict the users will select subsequent to each action; and

training the predictive model to predict the user'"'"'s behavior and generate a behavior vector for a given a feature vector.