Ex Parte Stobbs et al

Board of Patent Appeals and Interferences

Sep 15, 2011

10806307 (B.P.A.I. Sep. 15, 2011)

UNITED STATES PATENT AND TRADEMARK OFFICE
UNITED STATES DEPARTMENT OF COMMERCE
United States Patent and Trademark Office
Address: COMMISSIONER FOR PATENTS
P.O. Box 1450
Alexandria, Virginia 22313-1450
www.uspto.gov
APPLICATION NO. FILING DATE FIRST NAMED INVENTOR ATTORNEY DOCKET NO. CONFIRMATION NO.
10/806,307 03/22/2004 Gregory A. Stobbs 9305-002DVA 1838
27572 7590 09/16/2011
HARNESS, DICKEY & PIERCE, P.L.C.
P.O. BOX 828
BLOOMFIELD HILLS, MI 48303
EXAMINER
CORRIELUS, JEAN M
ART UNIT PAPER NUMBER
2162
MAIL DATE DELIVERY MODE
09/16/2011 PAPER
Please find below and/or attached an Office communication concerning this application or proceeding.
The time period for reply, if any, is set in the attached communication.
PTOL-90A (Rev. 04/07)

UNITED STATES PATENT AND TRADEMARK OFFICE
____________

BEFORE THE BOARD OF PATENT APPEALS
AND INTERFERENCES
____________

Ex parte GREGORY A. STOBBS and JOHN V. BIERNACKI
____________

Appeal 2009-010443
Application 10/806,307
Technology Center 2100
____________

Before JAY P. LUCAS, JOHN A. JEFFERY, and
DENISE M. POTHIER, Administrative Patent Judges.

POTHIER, Administrative Patent Judge.

DECISION ON APPEAL
Claims 11-20 have been rejected. See Final Rej. 3-6. Although the
Notice of Appeal is silent regarding which claims are appealed, the
Supplemental Appeal Brief unambiguously states that only claims 13-20 are
appealed. See Supp. App. Br. 3. Accordingly, we confine our decision to
those claims. See Ex parte Ghuman, 88 USPQ2d 1478 (BPAI 2008)
(precedential). Following our decision, the Examiner should cancel the
non-appealed claims. See id.; see also the Manual of Patent Examining
Procedure (MPEP) § 1215.03, Rev. 3, July 2010.
Appeal 2009-010443
Application 10/806,307

2
Appellants therefore appeal under 35 U.S.C. § 134(a) from the
Examiner’s rejection of claims 13-20. We have jurisdiction under 35 U.S.C.
§ 6(b). We reverse.

STATEMENT OF THE CASE
Appellants’ invention relates to a system for analyzing patents using
linguistics and other computer techniques. See generally Spec. ¶¶ 0002,
0007-09. Claim 13 (which includes the limitations of claim 11 also
included) is reproduced below with the key disputed limitation emphasized:
11. A computer-implemented patent portfolio analysis method
comprising:
providing user-prescribed categories which were specified
by a user;
retrieving a corpus of patent information from a database,
wherein the patent information is information from multiple patent
documents;
analyzing said patent information to generate a category
model corresponding to at least one of said user-prescribed
categories; and
applying said model against said patent information to select
from said patent information a subset that fits said model and
storing said subset in association with a label corresponding to said
at least one of said user prescribed categories in a computer-
readable dataset.

13. The method of claim 11 wherein said patent information
includes claim text information to be analyzed and wherein said
analyzing step includes:
defining an eigenspace representing a training population of
training claims each training claim having associated training
text;
Appeal 2009-010443
Application 10/806,307

3
representing at least a portion of said training claims in said
eigenspace and associating a predefined category with each
training claim in said eigenspace; and
projecting the claim text information to be analyzed into said
eigenspace and associating with said projected claim text the
predefined category of the training claim to which said projected
claim text is closest within the eigenspace.

The Examiner relies on the following as evidence of unpatentability:
Snyder US 6,038,561 Mar. 14, 2000
(filed Sept. 15, 1997)

Agrawal US 6,233,575 B1 May 15, 2001
(filed June 23, 1998)
Andrew McCallum & Kamal Nigam, Text Classification by Bootstrapping
with Keywords, EM and Shrinkage 1999 Ass’n. of Computational
Linguistics Unsupervised Learning in Nat. Lang. Processing Workshop 52-
58 (1999), available at http://aclweb.org/anthology-new/W/W99/W99-
0908.pdf (“McCallum”).

THE REJECTION
The Examiner rejected claims 13-20 under 35 U.S.C. § 103(a) as
unpatentable over Snyder, Agrawal, and McCallum. Ans. 4-7.1

THE OBVIOUSNESS REJECTION
Regarding claim 13, the Examiner finds that neither Snyder nor
Agrawal teaches the analyzing step includes defining or using an eigenspace.
Ans. 5. The Examiner relies on McCallum to teach using an eigenspace to

1 Throughout this opinion, we refer to (1) the Appeal Brief filed April 10,
2008; (2) the Supplemental Appeal Brief field April 28, 2008; (3) the
Examiner’s Answer mailed July 9, 2008; and (3) the Reply Brief filed
September 5, 2008.
Appeal 2009-010443
Application 10/806,307

4
provide enough training examples to classify a large collection of
documents. Ans. 5. Particularly, the Examiner finds the bootstrapping
technique coupled with hierarchical shrinkage and expectation maximization
discussed in McCallum is similar to Appellants’ description of (1) an
eigenspace defined by eigenvectors that form clusters of patents having
similar meaning, and (2) associating predefined categories for training
claims in an eigenspace (Ans. 5, 8-9).
Appellants argue that McCallum does not disclose or teach using
eigenspaces. App. Br. 15-16; Reply Br. 5-7. Appellants assert that
McCallum’s bootstrapping technique requires no label documents (App. Br.
15) and that McCallum is, at best, an analogy to the claimed invention
(Reply Br. 5).

ISSUE
Under § 103, has the Examiner erred in rejecting claim 13 by finding
that Snyder, Agrawal, and McCallum collectively would have taught or
suggested the analyzing step includes defining an eigenspace representing a
training population of training claims?

FINDINGS OF FACT (FF)
1. The VNR Concise Encyclopedia of Mathematics explains
eigenvalues, eigenvectors, and an eigenspace. “A number λ is called an
eigenvalue (or characteristics value) of a linear transformation A if there
exists a vector x ≠ 0 such that A(x) = λ · x. The vector x is then called an
eigenvector of the transformation A belonging to λ. The eigenvectors
Appeal 2009-010443
Application 10/806,307

5
belonging to λ together with the null vector form a subspace, called an
eigenspace of A”2 (some italics omitted).
2. McCallum discloses a text classification approach that requires no
labeled documents. The process uses a small set of keywords to assign
approximate or preliminary labels to some of the unlabeled documents by
term-matching or matching keywords provided by a human. These
preliminary labels are the starting point for a bootstrapping process that
learns a naive classifier using Expectations-Maximization (EM) and
hierarchy shrinkage. McCallum, Abstract; 52-54.
3. McCallum uses the preliminary labels and unlabeled data to
generate a naive Bayes classifier during the bootstrapping process. This
involves: initializing all λj’s (i.e., interpolation weights among cj’s described
as vectors (McCallum 56)); iterating the EM algorithm to normalize the λj’s
to the root of the class hierarchy; calculating the expectation of the class
labels or the probabilistic-weighted class labels for each document using the
classifier; and incrementing the new λj’s by attributing each word of the
hold-out data probabilistically to the ancestors of each class. The classifier
then predicts a class label for an unlabeled document. McCallum 54-56.
4. Snyder discusses producing a three dimensional (3-D) plot of top
scoring claims at step 132 and gives a 3-D projection of the results. The
scores plotted are used to identify documents most closely or proximately
related. Figure 8C, entitled “3D clustering of similar patents,” illustrates a
3D plot visualization of an analysis conducted on two sets of patents using
step 132. Scores based on the similarity of clusters of patents are plotted in

2 W. Gellert et al., The VNR Concise Encyclopedia of Mathematics 378
(1975) (“VNR”).
Appeal 2009-010443
Application 10/806,307

6
the 3-D framework with the graphical representation. Snyder, col. 5, ll.
28-30, 51-52; col. 18, ll. 41-47, 59-61; col. 24, ll. 49-62; Figs. 2B, 8C.
5. Snyder discloses a cross-comparison algorithm or matching to
generate an aggregate matching score for patent A, claim X versus patent B,
claim Y. In this processing, weights from the word vector analysis are
compared or used. Snyder, col. 16, l. 57 – col. 17, l. 16; Fig. 2A.

ANALYSIS
Based on the record before us, we find error in the Examiner’s
obviousness rejection of claim 13 which calls for, in pertinent part, the
analyzing step to include defining an eigenspace. Appellants have not
defined the term, “eigenspace.” See generally Specification. We therefore
construe this term using the ordinary and customary meaning. VNR defines
an eigenspace as particular subspace formed from the eigenvectors
belonging to a number or eigenvalue, λ, together with the null vector. FF 1.
Thus, the cited prior art needs to teach or suggest defining this particular
subspace in the step of analyzing the patent information to generate a
category model corresponding to a user-prescribed category as required by
claim 13.
As Appellants argue (App. Br. 15; Reply Br. 7) and as understood by
an ordinarily skilled artisan (see FF 1), McCallum does not teach or suggest
such an eigenspace. Granted, McCallum has some similarities to the
Appellants’ invention, including analyzing documents to place them into
categories or generate category models corresponding to user-prescribed
categories. For example, McCallum generates preliminary labels for some
documents by matching keywords provided by a human (see FF 2), and uses
Appeal 2009-010443
Application 10/806,307

7
probabilistic weights to predict labels for the unlabeled documents or to
generate a category model (see FF 3). McCallum also states that variable λj
is a vector. See id. Yet, McCallum does not teach or suggest that this vector
or any other vector (see id.) is an eigenvector that satisfies the conditions
which creates an eigenspace (see FF 1). Moreover, the Examiner has not
shown that an ordinarily skilled artisan would have recognized that
McCallum’ discussion suggests defining an eigenspace. Thus, despite the
Examiner’s statements that McCallum and Appellants’ invention are similar
(Ans. 5, 8-9), the Examiner has not adequately demonstrated that McCallum
teaches or suggests the analyzing step includes defining an eigenspace or a
particular subspace having eigenspace characteristic as recited.
Nether Snyder nor Agrawal cure this deficiency. See generally
Snyder and Agrawal. Notably, while not relied upon by the Examiner,
Snyder discusses plotting or clustering patent scores within a 3-D space to
analyze similar patents and identify documents most closely or proximally
related. See FF 4. Snyder also states the weights used in cross comparison
processing are vectors. See FF 5. However, like McCallum, Snyder does
not discuss that these vectors define an eigenspace (see FF 1). We therefore
find that the Examiner has not established a prima facie case that the Snyder,
Agrawal, and McCallum collectively teach the analyzing step includes
defining an eigenspace representing a training population of training claims
as recited and further projects claim text information into the recited
eigenspace and associates the projected claim text with a predefined
category that the text is closest within the recited eigenspace.
For the foregoing reasons, Appellants have persuaded us of error in
the obviousness rejection of: (1) claim 13; (2) independent claim 14 which
Appeal 2009-010443
Application 10/806,307

8
recite commensurate limitations; and (3) dependent claims 15-20 for similar
reasons.

CONCLUSION
The Examiner erred in rejecting claims 13-20 under § 103.
Following this decision, the Examiner should cancel claims 11 and 12.

DECISION
The Examiner’s decision rejecting claims 13-20 is reversed.

REVERSED

msc

Appeal 2009-010443
Application 10/806,307

9
APPENDIX

W. Gellert et al., The VNR Concise Encyclopedia of Mathematics 378 (1975)
(“VNR”).

Notice of References Cited

Application/Control No.

10/806,307
Applicant(s)/Patent Under
Reexamination
GREGORY A. STOBBS, ET AL.
Examiner

CORRIELUS, JEAN
Art Unit

2162

Page 1 of 1

U.S. PATENT DOCUMENTS

* Document Number Country Code-Number-Kind Code Date MM-YYYY

Name

Classification

A US-

B US-

C US-

D US-

E US-

F US-

G US-

H US-

I US-

J US-

K US-

L US-

M US-
FOREIGN PATENT DOCUMENTS

* Document Number Country Code-Number-Kind Code Date MM-YYYY

Country

Name

Classification

N

O

P

Q

R

S

T
NON-PATENT DOCUMENTS
* Include as applicable: Author, Title Date, Publisher, Edition or Volume, Pertinent Pages)

U

W. Gellert et al., The VNR Concise Encyclopedia of Mathematics 378 (1975).

V

W

X

*A copy of this reference is not being furnished with this Office action. (See MPEP § 707.05(a).)
Dates in MM-YYYY format are publication dates. Classifications may be US or foreign.

U.S. Patent and Trademark Office
PTO-892 (Rev. 01-2001) Notice of References Cited Part of Paper No.
THEVNR
CONCISE .
,ENCYCLOPEDIA OF
~M ATHEM ATICS SECOND! '"'I '"'I EDITION
W Gellert . S.Gottwald

M. Hellwich . H. Kastner· H. KOstner

Editors

K.A.Hirsch .H. Reichardt

Scientific Advisors

~ VAN NOSTRAND REINHOLD

~ __ New York

------""".I&i!j!!;iIillIi1",rl,n!!H~9;mUntiilm~\ll!n'i,~1I';'i'I!!11'3}:~F~~i1Gt'l1lmlilllF~',~rllJ1aD~
\.. ,..""ijll.u~l!J.l~.li:t~-.W1llMiuJtth..~~ll.tu,",Ar< ... Jtfli~l~\h.t~\t1;\,;~ : ~ '\
Tt
01
SE
E(
M
L(
b(
C'
m
G © VEB Bibliographisches Institut Leipzig, 1975
7( Mathematics at a Glance
First American Edition 1977 tr;

Second American Edition 1989

TI
Library of Congress Catalog Card Number 99-26992

re
ISBN 0-442-20590-2

e\

All rights reserved, No part of this work covered by the copyright

th
hereon may be reproduced or used in any form or by any means

8J graphic, electronic, or mechanical, including photocopying,

01
recording, taping, or information storage and retrieval systems ­
without written permission of the publisher.
fc

Made in the German Democratic Republic,
in

m Published by Van Nostrand Reinhold

115 Fifth Avenue
fr

New York, New York 10003

TI

Van Nostrand Reinhold International Company Limited
dl 11 New Fetter Lane

London EC4P 4EE, England

Van Nostrand Reinhold

480 La Trobe Street

Melbourne, Victoria 3000, Australia

E Macmillan of Canada

51 Division of Canada Publishing Corporation

164 Commander Boulevard
h
Agincourt, Ontario MIS 3C7, Canada
16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1u

s'

Ir

tt

Library of Congress Cataloging-in-Publication Data
T
n Main entry under title:

rr

n The VNR concise encyclopedia of mathematics,

First published under title: Mathematics at a glance,

Includes index,

I. Mathematics-Handbooks, manuals, etc, I. Gottwald, S,

\I, Van Nostrand Reinhold Company,

QA40V18 1989 510-dc19 88·26992
ISBN 0-442·20590·2
A
378 17. Linear algebra
Of particular importance are orthogonal matrices, because they transform orthonormal
into one another. Expressed in terms of coordinates this says that: The coordinates with
to one rectangular coordinate system are transformed into those with respect to another by
an orthogonal matrix. A matrix is orthogonal if AT = A-I. This equation can also be writte
the form A . AT = 1 and interpreted thus: In an orthogonal matrix the inner product of
rows is zero, the inner product of a row with itself is one. The same statements are tru
columns of A, and either set is a sufficient condition for the matrix to be orthogonal. For
every 2 x 2 orthogonal matrix can be written in the form:
COS 9' - si n 9') (COS 'P sin 9')
. or. .( sm !p cos !p sm 9' - cos 'P
In the first case the matrix represents a rotation of the plane through the angle
case there is an additional reflection in a line. Matrices of the second type can be distinguished
those of the first by the fact that their determinant is -1, whereas the determinant (
always +1. In general, the determinant of a orthogonal matrix is always +1 or -1. If
matrix has determinant +1, it is sometimes called proper, in general they correspond to
preserving orthogonal transformations of a Euclidean vector space. The following matrices
this type:
COS'P -sin'P 0) (COS'P 0 -Sln'P) (1 0 Q
A12(9') = sin'P cos 'P 0 , A 13('P) = 0 1 0 ,An(8-) = 0 cos (f -sin
( o 0 1 sin'P 0 cos'P 0 sinO
Here If. 'P, and () are arbitrary angles. If a fixed order of the basis vect;)rs el, e~, e3 is chosen,
A 12(11') represents a rotation of space about the eJ-axis. The eh ei-plane is rotated through If!
e3 is left unchanged. This fact gives rise to the special form of the matrix. Every proper
3 X 3 matrix A can be written as a product A = A13({)' A13('P)' A12(tp) for suitable
11', 'P, and {}.
Ju~t as for the orthogonal transformations, so the set of orthogonal n X n matrices

group. The proper orthogonal matrices form a subgroup of this group.

17.6. Eigenvalues
Eigenvalues and eigenvectors. A number A is called an eigenvalue (or characteristic value)
linear transformation A if there exists a vector x =4= 0 such that A(x) = A • x. The vector x is
called an eigenvector of the transformation A belonging to A. The eigenvectors belonging to ),
with the null vector form a subspace, called an eigenspace of A.
If the equation A(x) = ). . x is rewritten in the form (A - U) x = 0, then it can be stated
In this formulation it is possible to define an eigenvalue in terms of a matrix A
transformation A: A number A is an eigenvalue of the matrix A if A - ).T is singular.
Example 1: Let A be a singular transformation; then there exists a non-zero vector x

A(x) = 0 = 0.. x. Hence). = 0 is an eigenvalue of A, and the non-zero vectors of the

the <;igenvectors belonging to O.

Example 2: Suppose that the matrix A repre­ AI 0
A(xt) = At.XI _ 0 A2senting the operator A with respect to· a basis . . A-A- .
Xl ..... x. is diagonal:

Then the basis vectors are all eigenvectors of A(x.) = A.X. ( 0 0

A. Such transformations are particularly easy to

describe. because they change the basis vectors only by multiplying them by scalars.

called diagonal (or diagonalizable) transformations. Every transformation of an

space with n distinct eigenvalues is diagonalizable.

The significance of eigenvalues in physics. Eigenvalue problems are important in many
of physics. They make it possible to find coordinate systems in which the transformations'
take on their simplest forms. In mechanics for instance, the principal moments of a
are found with the help of the eigenvalues of the symmetric matrix representing its
The situation is similar in the mechanics of continua, where the rotations and deformation
body in the principal directions are found with the help of the eigenvalUes of a symmetric
Eigenvalues are of central importance in quantum mechanics, in which the measured values
17.6. Eigenvalues 379
'observables' appear as the eigenvalues of certain operators. The term 'transformation'
predominantly in pure mathematical (geometrical) context, whereas 'operator' is more
in applications (physics, technology).
,mputation of eigenvalues and eigenvectors. If a basis in a vector space V is chosen, then the
(A - AI) (x) = 0 is represented by the following system of equations for the coordinates
, ..., x. of x:
(all - A.) Xl + 012 X2 + ... + at, X. = 0
a2lxI + (a22 - A.) X2 + ... + 02, x. = 0· .· .· .
a,lxI + 0.2 X2 + ... + (a.. ).) X. = O.
coefficient matrix is the matrix A - U representing the transformation A - AI. Since only
vectors can be eigenvectors, the problem is to find non-zero solutions of this homogeneous
A necessary and sufficient condition for the existence of such solutions is that the deter­
of the matrix of coefficients should vanish: det (A - U) O. This is the case if and only
- AI is singular, that is, if). is an eigenvalue of A. The determinant can be seen to be a polynomial
degree n in A.:
det (A U) = ao + alA. + ... + a.).'.
is called the characteristic polynomial of the matrix A. If A' is another matrix representing A •
.4' = C-1AC for some matrix C, and its associated polynomial is the same:
find an eigenvector x one must therefore first find a root of the characteristic polynomial of A.
Jrdinates Xl. X2, ••• , x. of x can then be found as a non-trivial solution of the homogeneous
given above.
3: For n = 2 and A = ( 2 3) the eigenvalues are roots of the equation:
-1 -2

2 -). 3 1
det (A -).1) = = ).2 - I = O.
-1 -2 ).1
they are +1 and L The coordinates Xl, x 2 of the eigenvectors belonging to the eigenvalue
re the solutions of the system;
where T is an arbitrary non-zero number. In + 3X2 = 0::>---'-0 ' (XI.X2)=T·(-3,1), general, eigenvectors are determined only up
- 3x2 ­ to scalar multiples.
Tbe transformation to principal axes
symmetric transformations the theory leads to a particularly simple result. All the eigenvalues
5ymmetric transformation are real and there ex.ists an orthonormal basis of eigenvectors. If
represented by the matrix A, this means that there exists an orthogonal matrix C such that
C-1AC is diagonal, with the eigenvalues on the main diagonal. A' is called the normal form
and the change of basis represented by C is called the transformation to principal axes. The
C is the matrix of the coordinates of an orthonormal basis of eigenvectors with respect to
,is under which A is represented by the matrix A.
4: For A -+ A = ( 3 -1) the ei~llvalues are +2 and +4. The eigenvecto~ belong­
-1 3 '
(~1> X2) = Tl • (1, 1) and the eigenvectors belonging to +:4 ace (Xl' xi)':::: Tl(-l.l).
T1and T,2 can bechPSen to givC!the vectors the length 1. The eigenvectQfS,(1/V2, l/V2)
1/Y'2) form an orthonQrrnal ~is, and C is the matrix.
C =,,(1/1"2 -1/1"2). C- l =C1' = ( 1/V2 IIV2). '(2 0)C-1AC= 0 4
1/Y'2 1/V2' -1/Y'2 1/1"2'
means of the transformation to principal axes the equation of centred conics or quadrics
considerably simplified, by changing the Cartesian coordinate system to one consisting of
axes of the curve, or surface. These are the principal axes of the figure, which explains
transformation to principal axes.
5,,: ,Ifax2 +2bxy + cy2 = d is the equaHon of a conic section, then B]ab)
!ates on ,the left-hand side are arranged in a symmetric matrix A. A =
coordinate transformation to new rectangular coordinates (x', y'), by b c