Semantic analysis system for ranking search results

• US 10,846,295 B1
• Filed: 08/08/2019
• Issued: 11/24/2020
• Est. Priority Date: 08/08/2019
• Status: Active Grant

• Current Assignee: Applied Underwriters, Inc. (Berkshire Hathaway Incorporated)
• Original Assignee: Applied Underwriters, Inc. (Berkshire Hathaway Incorporated)
• Inventors: Clark, David Alan, Smith, Justin N., Nowotarski, Mark S.
• Primary Examiner(s): Jalil, Neveen Abel
• Assistant Examiner(s): Cheung, Hubert
• Application Number:
  US16/535,159
• Time in Patent Office: 474 Days
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/9035 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06F 16/9038 Presentation of query results
• CPC Class Codes: G06F 16/93 Document management systems

Abstract

A computer-implemented semantic analysis system for ranking search results has instructions including the steps: receive a query about a first document authored by a person; read in data identifying the first document; read in a second document about the person authored after the first document; identify indicator terms in the second document related to whether or not there is a significant discrepancy between the second document and the first document; read in weighting factors for the identified indicator terms; determine an aggregate weighting factor that decreases with an increasing number of indicator terms found; and calculate a score of how much of a discrepancy there is between the second document and the first document using the weighting factors. When multiple first documents are returned for multiple persons, present the links in order from high score to low score.

Claims

1. A semantic analysis system for ranking search results comprising:

a) an input device for reading in data;

b) a microprocessor for processing said data;

c) an output device for outputting said processed data; and

d) a permanent memory comprising computer readable instructions to physically cause said microprocessor to carry out the steps;

i) receive from a user via said input device, a search query;

ii) responsive to said search query, read in data identifying a first document authored by a person in a set of persons wherein;

1) said first document comprises textual content describing an injury suffered by said person on or before a first time;

2) said first document was authored at said first time; and

3) said first document is computer accessible by a link;

iii) read in a second document wherein said second document was authored after said first time and wherein said second document comprises textual content about an activity performed by said person after said first time;

iv) read in a set of indicator terms that have been statistically linked to a discrepancy between said textual content of said second document and said textual content of said first document wherein said discrepancy indicates that said person did not suffer said injury on or before said first time, and wherein said statistical linkage has been done without input from said first document, said second document, or said user;

v) determine which of said indicator terms are present in said second document;

vi) determine a score indicating how much of said discrepancy there is between said textual content of said second document and said textual content of said first document according to the steps;

1) read in an indicator weighting factor for each of said indicator terms present in said second document;

2) determine a first sum of all of said indicator weighting factors for all of said indicator terms present in said second document;

3) determine a second sum equal to said first sum plus a constant D;

4) determine a count N of the number of said indicator terms present in said second document; and

5) determine said score by multiplying said second sum by an aggregate weighting factor C_N that has a value of 1 when N is 1 and decreases with increasing N over a range of N;

vii) compare said score to a threshold; and

viii) provide by said output device said link to said first document when said score is greater than or equal to said threshold.

2. The semantic analysis system of claim 1 wherein said steps further comprise:

a) execute said steps ii to vii for each person in said set of persons to determine a score for each first document authored by said each of said persons, wherein each of said first documents is accessible by a link;

b) rank said links to said first documents in order of highest score to lowest score; and

c) output by said output device, said links to said first documents in said order of said rank.

3. The semantic analysis system of claim 1 wherein C_N decreases from 1.0 to 0.8 or less over a range of N of 1 to 2.

4. The semantic analysis system of claim 1 wherein said steps further comprise:

a) determine a first number by subtracting 1 from N;

b) determine a second number by multiplying said first number by a parameter A and then adding 1; and

c) determine said C_N by taking the inverse of said second number.

5. The semantic analysis system of claim 4 wherein said parameter A is about 0.07.

6. The semantic analysis system of claim 1 wherein D is set to a value less than zero.

7. The semantic analysis system of claim 1 wherein said first document is an application form submitted to a third party for approval.

8. The semantic analysis system of claim 7 wherein said first document is one or more of:

a) a job application;

b) an application for citizenship;

c) an insurance claim;

d) a posting on a dating web site;

e) an application for security clearance;

f) a loan application;

or

g) a patent application.

9. The semantic analysis system of claim 1 wherein said second document is one or more of:

a) a social media posting authored by said person;

or

b) a blog post authored by another person who is not said person.

10. The semantic analysis system of claim 1 wherein said steps further comprise:

a) determine a numerator equal to a square of said score;

b) determine a denominator equal to 1 plus said square of said score;

c) determine a probability of there being a significant discrepancy between said content of said second document and said content of said first document by taking a ratio of said numerator to said denominator;

d) determine a plurality of said probabilities for a plurality of first documents authored by a plurality of said persons in said set of persons wherein;

i) each of said plurality of said probabilities is based on a plurality of second documents each associated with each one of said plurality of said persons in said set of persons; and

ii) each of said second documents comprises content;

e) determine an average of said plurality of said probabilities; and

f) automatically adjust said threshold such that said average of said plurality of said probabilities has a desired value.

11. The semantic analysis system of claim 10 wherein said desired value of said average of said plurality of said probabilities is 0.8 or greater.

12. The semantic analysis system of claim 10 wherein said steps further comprise steps for statistically linking said indicator terms to said discrepancy between said textual content of said second document and said textual content of said first document, said steps for statistically linking comprising:

a) read in for each of said first documents associated with said plurality of persons, an adjudication of whether or not there is a significant discrepancy between each of said contents of said second documents and each of said contents of said first documents;

b) determine a plurality of residuals between each of said probabilities in said plurality of probabilities and each of said adjudications;

c) determine a sum of squares of said residuals;

d) automatically adjust said indicator weighting factors and said aggregate weighting factor such that said sum of squares of said residuals has a minimized value.

13. The semantic analysis system of claim 12 wherein said adjudications have a numerical value of 1 when there is a significant discrepancy and a numerical value of 0 when there is not a significant discrepancy.