US 7,734,633 B2Granted

Listwise ranking

AssigneeMicrosoft Corporation

InventorsTie-Yan Liu, Hang Li, Tao Qin, Zhe Cao

Life of this patent

Estimating the term…

Abstract

Procedures for learning and ranking items in a listwise manner are discussed. A listwise methodology may consider a ranked list, of individual items, as a specific permutation of the items being ranked. In implementations, a listwise loss function may be used in ranking items. A listwise loss function may be a metric which reflects the departure or disorder from an exemplary ranking for one or more sample listwise rankings used in learning. In this manner, the loss function may approximate the exemplary ranking for the plurality of items being ranked.

Claims17

Dependent claims

A method comprising:

deriving a listwise loss function from one or more sample listwise rankings, individual listwise rankings including individual items ordered based on the relevancy of the individual items, with respect to an exemplary listwise ranking;
and
using a computing device, ranking a plurality of items using the listwise loss function, wherein the listwise loss function is:
L
⁡
(
y
(
i
)
,
z
(
i
)
⁡
(
f
ω
)
)
=
-
∑
∀
g
∈
G
k
⁢
P
y
(
i
)
⁡
(
g
)
⁢
log
⁡
(
P
z
(
i
)
⁡
(
f
ω
)
⁡
(
g
)
)
;
in which:
y(i) is a judgement value for an item with respect to a query;
Py(i) is a probability of y(i);
z(i) is a list of scores for y(i) with a neural network model function ƒω applied to individual y(i) according to an equation
z
(
i
)
=
(
f
⁡
(
x
1
(
i
)
)
,
…
⁢
,
f
⁡
(
x
n
(
i
)
(
i
)
)
)
;
Pz(i)(ƒω) is a probability of z(i);
and
-
∑
∀
g
∈
G
k
⁢
P
y
(
i
)
⁡
(
g
)
⁢
log
⁡
(
P
z
(
i
)
⁡
(
f
ω
)
⁡
(
g
)
)
is a cross entropy for the query “q(i)” in which Gk is a top k subgroup for the query and g is an element of collection Gk.

132 more elements

Dependencies are inferred from the claim text.

patentgrep

Loading patent…

Abstract