System and method for personalized search

• US 8,903,811 B2
• Filed: 04/01/2009
• Issued: 12/02/2014
• Est. Priority Date: 04/01/2008
• Status: Active Grant

• Current Assignee: Certona Corporation (StartApp, Inc.)
• Original Assignee: Certona Corporation (StartApp, Inc.)
• Inventors: Hueter, Geoffrey J., Quandt, Steven C.
• Primary Examiner(s): Wong, Don
• Assistant Examiner(s): HEBERT, THEODORE E
• Application Number:
  US12/416,434
• Publication Number:
  US 20090248682A1
• Time in Patent Office: 2,071 Days
• US Class Current: 707/723
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 10/00 Administration; Management
• CPC Class Codes: G06Q 30/00 Commerce
• CPC Class Codes: G06Q 30/0185 Product, service or business identity fraud
• CPC Class Codes: G06Q 30/0201 Market modelling; Market analysis; Collecting market data
• CPC Class Codes: G06Q 30/0202 Market predictions or forecasting for commercial activities
• CPC Class Codes: G06Q 30/0204 Market segmentation
• CPC Class Codes: G06Q 30/0255 based on user history
• CPC Class Codes: G06Q 30/0269 based on user profile or attribute
• CPC Class Codes: G06Q 30/0631 Item recommendations
• CPC Class Codes: H04L 51/212 using filtering or selective blocking

Abstract

A system and method is disclosed for profiling a subject'"'"'s search engine keywords and results based on relevancy feedback. Because the system is based on the search behavior of the user, the profiling is language independent and balances the specificity of search terms against the profiled interests of the user. The system can also synthesize new keyword combinations to assist the user in refining the search or acquiring related content. The system has application in text mining, personalization, behavioral search, search engine optimization, and content acquisition, to name but a few applications.

Claims

1. A computer-implemented method comprising:

generating a score for a subject and search keyword objects of target object provided by an online computer-based search directed to an object type, either by explicit or by behaviorally inferred subject responses, wherein the generated object score is based on relevancy to the subject and search keywords of the target objects provided by the online computer-based search;

representing each of said subject and said objects in individualized vector form at a computer server, wherein the predicted relevancy of a target object to a search keyword object is generated by matching the search keyword object vector to the target object vector, and the predicted affinity of the subject to the target object is generated by matching the subject vector to the target object vector, further wherein said matching is calculated as the dot product between said vectors;

generating search results by matching profiles of said subject and keywords with profiles in a target object catalog and ranking against said profile of said subject and keywords; and

presenting said subject with top-ranking target objects, such that said top-ranking target objects are tailored to said subject;

wherein the presented target objects are restricted to the object type of the online computer-based search; and

wherein the computer server is further configured to generate the subject vector and the object vector by producing initial subject vectors and initial object vectors having respective initial dimensions, to determine predicted search relevance scores based on the initial subject vectors, initial object vectors, and subject search response data, and to calculate a cost function as the mean squared error between the predicted relevance scores and actual relevance scores across all said subject responses; and

wherein the computer server iteratively increases the dimensions of the generated subject and object vectors and generates the values of the added dimensions of both said subject and object vectors to reduce the cost function based on the differences between the predicted relevance scores and actual relevance scores, until the cost function decreases to a predetermined value, and wherein the actual relevance scores are based on user input.

2. The method according to claim 1, wherein said generating a score comprises causing said subject to score said relevancy of objects only by behavioral inference.

3. The method according, to claim 1, wherein said presenting said subject with top-ranking target objects includes presenting said top-ranking objects and content of different languages.

4. The method according to claim 1 wherein said presenting said subject with top-ranking target objects includes combining user behavioral search relevancy scores with textual search relevancy scores.

5. The method according to claim 1, wherein said step of representing said subject and said objects in vector form includes:

deriving said vector form solely from said subject'"'"'s relevancy ratings of said objects weighting relevancy ratings of users designated as expert, higher than relevancy ratings of users designated as non-experts and increasing said weighting for repeated occurrences of either viewing or not viewing a search result,

6. The method according to claim 1, further comprising clustering search objects in vector form to identify content regions of interest, to said subject.

9. The method according to claim 1, wherein further comprising updating said profile of said subject separately by decoupling relevancy scores and weights for said subject from other subjects, such as to train said subject vector individually, thereby enabling individualized targeting of increasing numbers of said subjects.

10. The method according to claim 1 wherein the step of representing said subject in vector form comprises including one or more of search interests or behavioral responses to search results as vector generation input data.

7. A method for personalized search, the method comprising the steps of:

submitting keywords to a search engine at a computer to generate keyword search results;

presenting said keyword search results from the computer to a user;

receiving user-specific search results at the computer;

rating relevancy of the user-specific search results according to user response;

developing or updating user, keyword, and content profiles, each of the respective profiles being in vector form;

updating the keyword profiles and content profiles, and displaying updated search results to the user, such that said user and each object or a group profile thereof are symmetric with each other and are each profiled through an individual vector such that the values and number of dimensions of said vectors are based solely on the user responses to the objects;

clustering and synthesizing an updated set of keywords;

generating target objects by matching a. profile of said user with profiles in a target object catalog and ranking against said profile of said user,

presenting said user with top-ranking target objects, such that said top-ranking target objects are tailored to said user,

wherein the presented target objects are restricted to the object type of the user-selected search result; and

further wherein the user vector, keyword vector, and the content vector each have a respective number of dimensions; and

further wherein the predicted relevancy rating of a content object to a search keyword is calculated by matching their object vectors; and

further wherein object vectors are generated by producing object vectors having respective initial dimensions, determining predicted relevancy ratings based on the initial object vectors, and calculating a cost function as the mean squared difference between the predicted relevancy ratings and the user relevancy ratings across all said user relevancy ratings; and

further wherein the dimensions of the generated object vectors are iteratively increased, and the values of the added dimensions are calculated to reduce said cost function, until said cost function reaches a predetermined value and wherein the user relevancy ratings are based on user response to search results; and

further wherein the user vectors are generated from the said content object vectors and said user relevancy ratings derived from said user response.

8. The method according to claim 7, wherein said step of submitting keywords includes the step of:

searching for products or content on a website.