Method for predicting ratings

• US 6,249,785 B1
• Filed: 05/06/1999
• Issued: 06/19/2001
• Est. Priority Date: 05/06/1999
• Status: Active Grant

• Current Assignee: Media Choice Inc.
• Original Assignee: Mediachoice Incorporated
• Inventors: Paepke, C. Owen
• Primary Examiner(s): Choules, Jack
• Assistant Examiner(s): BLACK, LINH
• Application Number:
  US09/306,550
• Time in Patent Office: 775 Days
• Field of Search: 707/5, 707/6, 707/7
• US Class Current: 1/1
• 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 707/99937 Sorting

Abstract

Predictions are based primarily upon similarities in pairs of ratings, irrespective of the actual value of the ratings. A table is used to translate each pair of ratings into rankings that are used to make predictions of future ratings. Similar ratings are ranked higher than dissimilar ratings. The prediction is based upon the average () of the books linked to the book of interest, as rated by the user, plus the difference (δ) between the average rating of the book of interest, as rated by all users, and the average ratings of the linked books, as rated by the user. The averages may be weighted by the rankings. Alternatively, the prediction is based upon the cumulative values applied to books linked to the books rated by the user, where the values are based upon the user'"'"'s ratings of the rated books.

Claims

1. A method for computerized prediction of a person'"'"'s rating of an object of art, said method comprising the steps of:

(a) storing rating data on n objects of art in at least one database in a computer;

(b) using the computer to generate up to _nC₂ pairs of ratings of the objects;

(c) using the computer to rank each pair of ratings according to similarity; and

(d) using the computer to predict a rating for an object of interest based upon the ratings of objects linked to the object of interest by at least one pair of ratings.

2. The method as set forth in claim 1 wherein said ranking step includes the step of:

(i) using a computer to convert each pair of ratings to a number corresponding to the similarity of the ratings.

3. The method as set forth in claim 2 wherein said ranking step further includes the step of:

(ii) using the computer to reduce the rank of infrequently occurring pairs.

4. The method as set forth in claim 3 wherein said ranking step further includes the step of:

(iii) using the computer to ignore pairs below a predetermined rank.

5. The method as set forth in claim 4 wherein said ranking step further includes the step of:

(iv) using the computer to quantize the remaining ranks.

6. The method as set forth in claim 1 wherein said predicting step includes the step of:

(i) using the computer to calculate the average rating, , of objects linked to the object of interest by a particular user;

(ii) using the computer to calculate the weighted average rating, y, by all users of objects linked to the object of interest by a particular user;

(iii) using the computer to calculate the average rating, b, of the object of interest by all users; and

(iv) using the computer to calculate (−

(b−

y)) as the predicted rating of the object of interest;

wherein $y=\frac{\sum <br><br>\left(\stackrel{\_}{r}·<br><br>\mathrm{\rho <br><br>}\right)}{\sum <br><br>\mathrm{\rho <br><br>}}$

and ρ

is the ranking of the pairs of ratings.

7. The method as set forth in claim 6 wherein said predicting step includes the step of:

(v) using the computer to round off the predicted rating to an integer.

8. The method as set forth in claim 6 wherein ρ

is a weighted, quantized representation of the rankings of selected pairs of ratings of objects, including the object of interest and the linked objects.

9. The method as set forth in claim 1 wherein said predicting step includes the step of using the computer to exclude objects based upon demographic data.

10. The method as set forth in claim 1 wherein said predicting step includes the step of using the computer to exclude objects based upon predetermined search criteria.

11. The method as set forth in claim 1 wherein said predicting step includes the steps of:

(i) using the computer to build a table of values by applying values to objects linked to the objects rated by the person; and

(ii) displaying at least the object associated with the highest value.

12. The method as set forth in claim 9 wherein said displaying step is preceded by the step of sorting the values in the table.

13. A computerized method for predicting a person'"'"'s rating of an object of art, said method comprising the steps of:

(a) storing rating data on n objects of art in at least one database in a computer;

(b) generating up to _nC₂ pairs of ratings of the objects;

(c) ranking each pair of ratings according to similarity;

(d) eliminating the pairs below a predetermined rank;

(e) predicting a rating for an object of interest based upon the remaining pairs.

14. The method as set forth in claim 13 wherein step (d) is followed by the step of:

eliminating the pairs that occur only once.

15. The method as set forth in claim 13 wherein said prediction is based upon the average rating of the objects, as rated by the user, linked to the object of interest plus the difference (δ

) between the average rating of the object of interest, as rated by all users, minus the average ratings of the linked objects, as rated by all users in the database.

16. The method as set forth in claim 13 wherein said prediction is based upon the average of values applied to objects linked to the object of interest, wherein the values are based upon ratings of the linked objects by the user inquiring about the object of interest.