James Mayfield TAC 2013 Entity Linking · 2014. 6. 4. · TAC 2013 Entity Linking English,...
Transcript of James Mayfield TAC 2013 Entity Linking · 2014. 6. 4. · TAC 2013 Entity Linking English,...
TAC 2013Entity Linking
English, Spanish, Chinese
James Mayfield
Special thanks to:Hoa Trang Dang
Heng Ji
Monolingual Entity Linking
And talking to the Today programme's John Humphrys, Scottish MP Michael Moore argued that being part of the UK provides Scotland with security. "When two of our biggest banks - RBS and Bank of Scotland - collapsed, it was the strength and size of the UK economy that helped us to cope."!
Dentist Michael Moore of Boaz Alabama was seen last Wednesday at the home of Penelope Binkels
Cross-Language Entity Linking�(�*0�1&'��0��.�-;����7��4�9.��03"2�;�������别# �%�)��.�9���� 08,���=!�4��6��/*�+)�9!�0�.5��=�0��(迈 尔<$尔:
Sample English Entity Linking Queries
<query id="EL13_ENG_0277"> <docid>LTW_ENG_20091010.0098</docid> <name>UCLA</name> <beg>612</beg> <end>615</end> </query>
<query id="EL13_ENG_1207"> <docid>AFP_ENG_20100325.0138</docid> <name>Taylor Phinney</name> <beg>107</beg> <end>120</end> </query>
<query id="EL13_ENG_1253"> <docid>bolt-eng-DF-170-181137-9030230</docid> <name>Motor City</name> <beg>113360</beg> <end>113369</end> </query>
2190 Queries
1100 NILs 1090 Linked
Joe Ellis
Scoring Metric: B-Cubed+
A clustering has high B-Cubed recall if, given an item, most of its coreferential items are in its cluster
A clustering has high B-Cubed precision if, given an item, most of the other items in its cluster are in fact coreferential with it
B-Cubed+ adds the constraint that links to existing items in the KB must also match
English Entity Linking Results - All Runs
0.000#
0.100#
0.200#
0.300#
0.400#
0.500#
0.600#
0.700#
0.800#
MS_MLI2#
SYDN
EY_C
MCR
C2#
THUNLP5#
UI_CC
G5#
HITS1#
UI_CC
G1#
HITS3#
THUNLP1#
RPI_BLEN
DER6
#basistech1#
basistech3#
UBC
7#MSRA_
Link2#
UBC
4#UMass_CIIR5#
UGE
NT_IBCN
4#UBC
8#INESCID3
#lkd1
#RP
I_BLEN
DER2
#RP
I_BLEN
DER7
#CA
SIA4
#lkd8
#CA
SIA2
#lkd9
#po
seidon
1#BU
PTTeam
3#lkd2
#USFD1
#BU
PTTeam
1#RP
I_BLEN
DER3
#OSU
4#ZZ_INFO
_TEC
H1#
Web
Sail1#
UMass_CIIR1#
TRRD
1#hltcoe
1#
B"Cubed+)F1)(All)"")2190)queries))
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"MS_MLI2"
SYDN
EY_C
MCR
C1"
THUNLP4"
UI_CC
G5"
HITS1"
RPI_BLEN
DER1
"UMass_CIIR4"
UBC
2"basistech5"
MSRA_
Link4"
UGE
NT_IBCN
1"UWashington1
"INESCID4
"po
lymtl5
"lkd1
"CA
SIA5
"po
seidon
1"BU
PTTeam
4"USFD2
"ZZ_INFO
_TEC
H4"
PRIS20132"
OSU
3"Web
Sail4"
TALP_U
PC1"
TRRD
2"hltcoe
1"Brande
is1"
B"Cubed+)F1)All)
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
MS_MLI2"
UBC10"
SYDNEY_CMCRC1"
THUNLP4"
INESCID4"
UI_CCG5"
MSRA_Link1"
HITS1"
UMass_CIIR4"
OSU3"
B"Cubed+)F1)In)KB)
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
MS_MLI1"
RPI_BLENDER1"
SYDNEY_CMCRC1"
UWashington1"
basistech4"
hltcoe1"
UI_CCG4"
THUNLP4"
UMass_CIIR5"
B"Cubed(F1(Not(In(KB(
0"0.1"0.2"0.3"0.4"0.5"0.6"0.7"0.8"
FRDC1" basistech5" THUNLP4" HITS3"
B"Cubed+)F1))All)
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
FRDC2" THUNLP2" basistech4" HITS1"
B"Cubed+)F1)In)KB)
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
basistech5" THUNLP4" HITS3" FRDC3"
B"Cubed+)F1)Not)In)KB)
Chinese
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
HITS2" basistech2" INESCID2"
B"Cubed+)F1)All)
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
HITS2" basistech2" INESCID2"
B"Cubed+)F1)In)KB)
0"
0.1"
0.2"
0.3"
0.4"
0.5"
0.6"
0.7"
0.8"
basistech5" HITS2" INESCID1"
B"Cubed+)F1)Not)In)KB)
Spanish