Ranking using word overlap and correlation features
A system for and method for ranking results. The system includes a server configured to receive a query and an advertisement engine configured to receive the query from the server. The advertisement engine ranks advertisements based on various features, including at least one word overlap feature and a correlation feature.
1. A system for ranking sponsored results, the system comprising:
a server configured to receive a query;
an advertisement engine configured to receive the query from the server, the advertisement engine ranking an advertisement of a plurality of advertisements based on a plurality of features, the advertisement engine calculating at least one word overlap feature between each advertisement and the query, wherein the plurality of features include at least one word overlap feature and a correlation feature; and
wherein the at least one word overlap feature includes a first feature that identifies whether all terms of the query are present in the advertisement, a second feature that identifies whether some of the terms of the query are present in the advertisement, a third feature that identifies whether none of the terms of the query are present in the advertisement, and a fourth feature that identifies a percentage of terms of the query present in the advertisement.
2. The system according to claim 1 , wherein the correlation feature comprises a Chi-squared characteristic.
3. The system according to claim 2 , wherein the Chi-squared characteristic is calculated according to the relationship:
x
2
=
L
(
o
11
o
22
-
o
12
o
21
)
2
(
o
11
+
o
12
)
(
o
11
+
o
21
)
(
o
12
+
o
22
)
(
o
21
+
o
22
)
where |L| is the number of queries in a query log, o 11 stands for a first number of queries in the query log, which contain both t 1 and t 2 , o 12 stands for a second number of queries in the query log, in which t 2 occurs but t 1 does not, o 21 stands for a third number of queries in the query log, in which t 1 occurs but t 2 does not, o 22 stands for a fourth number of queries in the query log, in which neither t 1 or t 2 occur.
4. The system according to claim 1 , wherein the correlation feature comprises a pointwise mutual information characteristic.
5. The system according to claim 4 , wherein the pointwise mutual information characteristic is calculated according to the relationship:
PMI
(
t
1
,
t
2
)
=
log
2
P
(
t
1
,
t
2
)
P
(
t
1
)
P
(
t
2
)
where t 1 is a term from the query, and t 2 is a bidded term from the advertisement, P(t) is the probability that term t appears in the query, and P(t 1 , t 2 ) is the probability that terms t 1 and t 2 occur in the query.
6. A method for ranking results, the method comprising:
receiving a query;
accessing a plurality of elements to be ranked;
calculating at least one word overlap feature between each element and the query, wherein the at least one word overlap feature includes a first feature that identifies whether all terms of the query are present in an element, a second feature that identifies whether some of the terms of the query are present in the element, a third feature that identifies whether none of the terms of the query are present in the advertisement, and a fourth feature that identifies a percentage of terms of the query present in the element, the word overlap feature being calculated on a computer;
calculating a correlation feature between each element and the query; and
ranking each element of the plurality of elements based on the word overlap feature and the correlation feature.
7. The method according to claim 6 , wherein the correlation feature comprises a Chi-squared characteristic.
8. The method according to claim 7 , wherein the Chi-squared characteristic is calculated according to the relationship:
x
2
=
L
(
o
11
o
22
-
o
12
o
21
)
2
(
o
11
+
o
12
)
(
o
11
+
o
21
)
(
o
12
+
o
22
)
(
o
21
+
o
22
)
where |L| is the number of queries in a query log, o 11 stands for a first number of queries in the query log, which contain both t 1 and t 2 , o 12 stands for a second number of queries in the query log, in which t 2 occurs but t 1 does not, o 21 stands for a third number of queries in the query log, in which t 1 occurs but t 2 does not, o 22 stands for a fourth number of queries in the query log, in which neither t 1 or t 2 occur.
9. The method according to claim 6 , wherein the correlation feature comprises a pointwise mutual information characteristic.
10. The method according to claim 9 , wherein the pointwise mutual information characteristic is calculated according to the relationship:
PMI
(
t
1
,
t
2
)
=
log
2
P
(
t
1
,
t
2
)
P
(
t
1
)
P
(
t
2
)
where t 1 is a term from the query, and t 2 is a bidded term from an ad, P(t) is the probability that term t appears in the query, and P(t 1 , t 2 ) is the probability that terms t 1 and t 2 occur in the query.
11. A non-transitory computer readable medium having stored therein instructions executable by a programmed processor for ranking results, the non-transitory computer readable medium comprising instructions for:
receiving a query;
accessing a plurality of elements to be ranked;
calculating at least one word overlap feature between each element and the query, wherein the at least one word overlap feature includes a first feature that identifies whether all terms of the query are present in an element, a second feature that identifies whether some of the terms of the query are present in the element, a third feature that identifies whether none of the terms of the query are present in the element, and a fourth feature that identifies a percentage of terms of the query present in the element;
calculating a correlation feature between each element and the query; and
ranking each element of the plurality of elements based on the word overlap feature and the correlation feature.
12. The computer readable medium according to claim 11 , wherein the correlation feature comprises a Chi-squared characteristic calculated according to the relationship:
x
2
=
L
(
o
11
o
22
-
o
12
o
21
)
2
(
o
11
+
o
12
)
(
o
11
+
o
21
)
(
o
12
+
o
22
)
(
o
21
+
o
22
)
where |L| is the number of queries in a query log, o 11 stands for a first number of queries in the query log, which contain both t 1 and t 2 , o 12 stands for a second number of queries in the query log, in which t 2 occurs but t 1 does not, o 21 stands for a third number of queries in the query log, in which t 1 occurs but t 2 does not, o 22 stands for a fourth number of queries in the query log, in which neither t 1 or t 2 occur.
13. The computer readable medium according to claim 11 , wherein the correlation feature comprises a pointwise mutual information characteristic calculated according to the relationship:
PMI
(
t
1
,
t
2
)
=
log
2
P
(
t
1
,
t
2
)
P
(
t
1
)
P
(
t
2
)
where t 1 is a term from the query, and t 2 is a bidded term from an ad, P(t) is the probability that term t appears in the query, and P(t 1 , t 2 ) is the probability that terms t 1 and t 2 occur in the query.
14. A system for ranking sponsored results, the system comprising:
a server configured to receive a query;
an advertisement engine configured to receive the query from the server, the advertisement engine ranking an advertisement of a plurality of advertisements based on a plurality of features, the advertisement engine calculating at least one word overlap feature between each advertisement and the query, wherein the plurality of features include at least one word overlap feature and a correlation feature; and
wherein the at the correlation feature is calculated based on a query term and a bid term, where the correlation feature is further calculated based on a first number of queries in the query log, which contain both the query term and the bid term, a second number of queries in the query log, in which the bid term occurs but the query term does not, a third number of queries in the query log, in which the query term occurs but the bid term does not, and a fourth number of queries in the query log, in which neither the bid term or the query term occur.
15. A system for ranking sponsored results, the system comprising:
a server configured to receive a query;
an advertisement engine configured to receive the query from the server, the advertisement engine ranking an advertisement of a plurality of advertisements based on a plurality of features, the advertisement engine calculating at least one word overlap feature between each advertisement and the query, wherein the plurality of features include at least one word overlap feature and a correlation feature; and
wherein the at the correlation feature is calculated based on a query term and a bid term, where the correlation feature is further calculated based on a probability function of whether the query term and the bid term occur in the same query.