# US 2013/0159323 A1 · Recommending tags based on user ratings

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2013/0159323 A1
- Document ID: US20130159323A1
- Status: Published application
- Kind: A1
- Application number: 13/706,727
- Filed: 2012-12-06
- Published: 2013-06-20
- Earliest claimed: 2011-12-16
- Assignees: International Business Machines Corporation
- Inventors: Sheng Hua; Xian Wu; Quan Yuan; Shiwan Zhao
- Web page: https://patentgrep.com/patent/US20130159323A1

## CPC

- G06F 7/24: Sorting, i.e. extracting data from one or more carriers, rearranging the data in numerical or other ordered sequence, and rerecording the sorted data on the original carrier or on a different carrier or set of carriers {sorting methods in general}(G06F7/36 takes precedence)
- G06F 16/38: Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A method of recommending tags for user includes receiving a current rating of current user on current objects; determining a candidate tag set based on the current user and the current rating; ordering each candidate tag in the candidate tag set based on index value thereof; and recommending tags for the current user based on the ordering.

## Claims

### Claim 1

A method of recommending tags for user, the method comprising:
receiving, with a processing device, a current rating of a current user on current objects;
determining a candidate tag set based on the current user and the current rating;
ordering candidate tags in the candidate tag set based on index values thereof; and
recommending tags for the current user based on the ordering.

### Claim 2 (depends on 1)

The method according to claim 1, wherein, the determining a candidate tag set based on the current user and the current rating comprises: determining history tags of the current user under the current rating as the candidate tag set based on predetermined correlative relationship of the user with history tags under different rating levels.

### Claim 3 (depends on 2)

The method according to claim 2, wherein, the correlative relationship of the user with content of history tags under different rating levels comprises similarity between each two of the objects, which is determined based on content similarity of history tags of the objects and rating similarity of different users on a same object.

### Claim 4 (depends on 3)

The method according to claim 3, wherein each object has a first vector and a second vector, the first vector is term frequency/inverse document frequency value of that object on corresponding history tags, and the second vector is history rating on that object, and determination of the similarity between each two of the objects comprises:
determining the content similarity Sim1(i, j) based on similarity between the first vectors;
determining the rating similarity Sim2(i, j) based on similarity between the second vectors; and
determining the similarity between each two of the objects Sim(i, j)=Sim1(i, j)*a+Sim2(i, j)*(1−a), wherein 0<a<1, parameter a is used for adjusting weight.

### Claim 5 (depends on 4)

The method according to claim 4, wherein, the determining history tags of the current user under the current rating as the candidate tag set based on predetermined correlative relationship of the user with history tags under different rating levels comprises:
determining K objects, which are tags that are determined based on the similarity between each two of the objects, have the highest similarity with the current objects and have been used by the current user; and
taking tags used by that current user on that K objects as the candidate tag set.

### Claim 6 (depends on 5)

The method according to claim 5, wherein determining K objects comprises further defining a rating of the current user as r.

### Claim 7 (depends on 5)

The method according to claim 5, wherein, the index values comprise weighted value
∑
k
=
1
K

v
t
*
Sim

(
i
,
k
)
of the candidate tags, wherein, if the current user uses tag t for k-th object, then vt=1, otherwise vt=0.

### Claim 8 (depends on 2)

The method according to claim 2, wherein the correlative relationship of the user with history tags under different rating levels comprises a tag generative probability p(t|u, r) of the user under different rating levels, and the index values comprise a generative probability of the candidate tags.

### Claim 9 (depends on 8)

The method according to claim 8, further comprising:
determining tag attribute distribution p(Xt), the tag attribute comprises objective and subjective;
determining distribution of object i on tag t p(t|i); and
determining the
p

(
t
|
u
,
r
)
=
∑
k
=
1
K

p

(
t
|
z
=
k
)

p

(
z
=
k
|
u
,
r
)
,
wherein K is the number of topics z of the history tags, p(t|z) is distribution of the topics z on the history tags, and the p(z|u, r) is distribution of users on topics z of the history tags under different rating levels.

### Claim 10 (depends on 9)

The method according to claim 9, wherein:
determining probability p(Xt=0) that each of the candidate tags is an objective tag and probability p(Xt=1) that each of the candidate tags is a subjective tag based on the tag attribute distribution p(Xt); and
computing generative probability of the candidate tags based on the equation p(Xt=0)*p(t|i)+p(Xt=1)*p(t|u, r).

### Claim 11

An apparatus for recommending tags for user, the apparatus comprising:
a rating means configured to receive a current rating of a current user on current objects;
a determining means configured to determine a candidate tag set based on the current user and the current rating;
an ordering means configured to order candidate tags in the candidate tag set based on index values thereof; and
a recommending means configured to recommend tags for the current user based on the ordering.

### Claim 12 (depends on 11)

The apparatus according to claim 11, wherein the determining means comprises: means configured to determine history tags of the current user under the current rating as the candidate tag set based on predetermined correlative relationship of the user with history tags under different rating levels.

### Claim 13 (depends on 12)

The apparatus according to claim 12, wherein the correlative relationship of the user with content of history tags under different rating levels comprises similarity between each two of the objects, which is determined based on content similarity of history tags of the objects and rating similarity of different users on a same object.

### Claim 14 (depends on 13)

The apparatus according to claim 13, wherein each object has a first vector and a second vector, the first vector is term frequency/inverse document frequency value of that object on corresponding history tags, and the second vector is history rating on that object, wherein the apparatus further comprises:
means configured to determine the content similarity Sim1(i, j) based on similarity between the first vectors;
means configured to determine the rating similarity Sim2(i, j) based on similarity between the second vectors; and
means configured to determine the similarity between each two of the objects Sim(i, j)=Sim1(i, j)*a+Sim2(i, j)*(1−a), wherein 0<a<1, parameter a is used for adjusting weight.

### Claim 15 (depends on 14)

The apparatus according to claim 14, wherein, the determining means further comprises:
means configured to determine K objects, which are tags that are determined based on the similarity between each two of the objects, have the highest similarity with the current objects and have been used by the current user; and
means configured to take tags used by that current user on that K objects as the candidate tag set.

### Claim 16 (depends on 15)

The apparatus according to claim 15, wherein, the K objects further define rating of the current user as r, too.

### Claim 17 (depends on 15)

The apparatus according to claim 15, wherein the index values comprise weighted value
∑
k
=
1
K

v
t
*
Sim

(
i
,
k
)
,
wherein, if the current user uses tag t for k-th object, then vt=1, otherwise vt=0.

### Claim 18 (depends on 12)

The apparatus according to claim 12, the correlative relationship of the user with history tags under different rating levels comprises tag generative probability p(t|u, r) of the user under different rating levels, and the index values comprise generative probability of the candidate tags.

### Claim 19 (depends on 18)

The apparatus according to claim 18, further comprising:
means configured to determine tag attribute distribution p(Xt), the tag attribute comprises objective and subjective;
means configured to determine conditional probability p(t|i) of generating the candidate tag t based on the object i; and
means configured to determine the
p

(
t
|
u
,
r
)
=
∑
k
=
1
K

p

(
t
|
z
=
k
)

p

(
z
=
k
|
u
,
r
)
,
wherein, K is the number of topics z of the history tags, p(t|z) is distribution of the topics on the history tags, and the p(z|u, r) is distribution of users on topics under different rating levels.

### Claim 20 (depends on 19)

The apparatus according to claim 19, comprising:
means configured to determine probability p(Xt=0) that each of the candidate tags is an objective tag and probability p(Xt=1) that each of the candidate tags is a subjective tag based on the tag attribute distribution p(Xt); and
means configured to compute generative probability of the candidate tags based on the equation p(Xt=0)*p(t|i)+p(Xt=1)*p(t|u, r).
