Granted Patent B2

US 9,317,584 · App. 13/731,069 · Granted Apr 19, 2016

Abstract

Claims (73)

1. A computer implemented web-based system for personalized search using keyword index pruning, comprising:

a computer implemented web-based service system, including a memory component and a processor component, for searching, collecting relevancy feedback, and acquiring content by keywords;

wherein profiles of subjects, including search users, are stored in a subject profile database and profiles of objects, including keyword and target content items, are stored in an object catalog databases;

said subject and keyword profiles are matched against target profiles and ranked to be optimal to keywords;

wherein optimal subject and object profiles are derived from said subject's relevancy ratings of search result objects against inputted keywords; and

further wherein said web-based service system processes, identifies and eliminates ineffective keywords by performing a token score check by operation of a behavioral engine wherein the input to the token score check is a set of behavioral scores corresponding to target objects and the behavioral scores are quantized into a discrete set of score bins x i , where x i denotes a contiguous set of behavioral scores, and further wherein the effectiveness of the keyword is measured by the entropy H of the distribution of behavioral scores; whereby said entropy is defined as

H

=

-

∑

i

=

1

n

⁢

p

⁡

(

x

i

)

⁢

log

⁢

⁢

p

⁡

(

x

i

)

where p(x i ) denotes the probability that the behavioral score of a particular target object is in the set x i .

2. The web-based system for personalized search using keyword index pruning, according to claim 1 , wherein said subject and keyword profiles are matched against target profiles and ranked to be optimal to keywords further includes a target object catalog database and matching and ranking of one subject or object type against a set of target objects of another type derived from said target object catalog database.

3. The web-based system for personalized search using keyword index pruning, according to claim 2 , wherein said target objects are restricted to those that are of the same type of object being requested in the keyword search.

4. The web-based system for personalized search using keyword index pruning, according to claim 1 , wherein said subject and keyword profiles are matched against target profiles and ranked to be optimal to keywords further includes combining the matching scores of the keyword profiles against the target object catalog with the matching scores of the subject profiles against the same target object catalog, whereby the search results are then ranked by the combined score.

5. The web-based system for personalized search using keyword index pruning, according to claim 4 , wherein said matching scores of the keyword profiles against the target object catalog includes breaking keyword phases into one or more keyword token objects, whereby the matching score of the keyword phrase against a target object is a combination of the matching scores of the individual keyword tokens.

6. The web-based system for personalized search using keyword index pruning, according to claim 5 , wherein said breaking keyword phases into one or more keyword token objects and matching score of the keyword phrase to a target object is a combination of the matching scores of the individual keyword tokens includes:

each token is considered separately;

for each token, the behavioral engine matches the token profile and the target object profiles to produce a behavioral score for each target object;

and the behavioral scores corresponding to a token are checked to see if they are sufficient to differentiate high relevance results from low relevance results;

whereby if the behavioral engine does not have a profile for a token, then the token is skipped, and for each token that passes the token score check, the corresponding target object behavioral scores are combined to form a final set of behavioral scores.

7. The web-based system for personalized search using keyword index pruning, according to claim 1 , wherein said keywords are processed by both a conventional text-based search engine, and a behavioral search engine that utilizes a behavioral profiling and targeting system whereby

a text-based search engine generates a relevancy score for each target object;

a behavioral search engine matches the subject and keyword profiles and the target object profiles to produce a behavioral score for each target object.

8. The web-based system for personalized search using keyword index pruning, according to claim 7 , wherein said behavior score is calculated as a combination of the scores that result from matching keyword objects with the target objects and the scores that result from matching the subject profile to the target objects;

whereby said behavior score is derived, and the relevancy scores and the behavioral scores are combined according to the convex combination:

COMBINED SCORE= a *(TEXT SCORE)+(1− a )*(BEHAVIOR SCORE),

where a is a continuous value between 0 and 1, inclusive, that determines the weighting or balance between the two search engines.

9. The web-based system for personalized search using keyword index pruning, according to claim 1 , wherein said web-based system for personalized search using keyword index pruning is language-independent, whereby said system does not parse or analyze the keywords or the target objects and provides a translation of keywords as an equivalence of search response as different language users perform searches for similar content.

10. The web-based system for personalized search using keyword index pruning, according to claim 1 , further comprising a behavior engine wherein when presented with an unknown keyword phrase, said behavior engine can bootstrap itself by presenting random results and then learning from a user response.

11. The web-based system for personalized search using keyword index pruning, according to claim 1 , further comprising a behavior engine wherein when presented with an unknown keyword phrase, said behavior engine can bootstrap itself by presenting results from matching the subject profile to the target object catalog and then learning from a user response.

12. The web-based system for personalized search using keyword index pruning, according to claim 1 , further comprising when a behavioral engine is used in a non-expert, general search environment:

the relevancy scores and weighting are derived using a target model based on user response to search result objects;

results that are presented and not selected by the user for further investigation are considered low-relevance and low-weight;

if a result is presented multiple times to the user and not selected, then that increases the weighting of that negative response; and

results that are selected are considered higher relevance and higher weight

whereby of the selected results further relevance can be attributed based on re-selecting a search result or spending more time examining the contents of a search result.

13. The web-based system for personalized search using keyword index pruning, according to claim 1 , further comprising the behavioral engine then evaluating whether the entropy H of the distribution of behavioral scores is above a certain threshold, thereby indicating lack of search discrimination; and eliminating keywords from the keyword profile index database that are greater than such entropy threshold.

14. The web-based system for personalized search using keyword index pruning, according to claim 1 , further comprising operating the system as a content acquisition engine wherein a search user utilizes the system to:

request and evaluate search results through interaction with a search result viewer;

generate relevancy feedback through interaction with the results in the results viewer;

generated relevancy feedback is used by the profiler to create profiles of the search users, keywords, and result objects; and

said profiles are then matched by the keyword synthesizer against other profiled keywords to create new trial keyword phrases.

15. The web-based system for personalized search using keyword index pruning, according to claim 14 , further comprising said system configured to:

submit an initiating keyword phrase by the search user to an external search engine, which then returns search results for review by the search user;

collect from the user implicit or explicit scores based upon the relevancy of the results;

profile the search user, keywords, and results based on the search interests and the associated responses of keywords and results;

eliminate the keyword tokens that do not meet the effectiveness criteria;

generate new keywords by matching the search user subject profile to the keyword profiles;

wherein the top-matching keyword tokens to generate new trial keyword phrases for subsequent search; and the process is iterated and refined as the search user provides feedback on the subsequent search results.

16. The web-based system for personalized search using keyword index pruning, according to claim 15 , wherein subject and keyword profiles are represented as vectors, whereby the matching of a subject or keyword vector to a target object vector generates a relevancy score.

17. The web-based system for personalized search using keyword index pruning, according to claim 16 , further wherein new content of interest is tagged as it comes into the system by setting a region of interest around a keyword profile vector.

18. The web-based system for personalized search using keyword index pruning, according to claim 17 , further wherein the user can adjust how much content is tagged by adjusting the size of the region of interest.

Assignments (10)