
计 算 机 研 究 与 发 展
DOI
:
10.7544∕issn1000G1239.2021.20190749
JournalofCom
p
uterResearchandDevelo
p
ment 58
(
8
):
1811 1819
,
2021
收稿日期
:
2019
-
10
-
23
;
修回日期
:
2020
-
09
-
07
基金项目
:
国家杰出青年科学基金项目
(
61425016
);
国家自然科学基金重大项目
(
91746301
);
泰山学者工程专项经费
(
ts201511082
)
Thisworkwassu
pp
ortedb
y
theNationalNaturalScienceFoundationofChinaforDistin
g
uishedYoun
g
Scholars
(
61425016
),
the
Ma
j
orPro
g
ramofNationalNaturalScienceFoundationofChina
(
91746301
),
andtheTaishanScholarsPro
g
ram ofShandon
g
ProvinceofChina
(
ts201511082
)
.
舆情场景下基于层次知识的话题推荐方法
史 存 会
1
,
3
胡 耀 康
2
,
3
冯
彬
1
,
3
张
瑾
2
俞 晓 明
1
刘
悦
1
程 学 旗
4
1
(
中国科学院计算技术研究所数据智能系统研究中心
北京
100190
)
2
(
中国科学院网络数据科学与技术重点实验室
(
中国科学院计算技术研究所
)
北京
100190
)
3
(
中国科学院大学
北京
100049
)
4
(
烟台中科网络技术研究所
山东烟台
264005
)
(
shicunhui@ict.ac.cn
)
A HierarchicalKnowled
g
eBasedTo
p
icRecommendation MethodinPublic
O
p
inionScenario
ShiCunhui
1
,
3
,
HuYaokan
g
2
,
3
,
Fen
g
Bin
1
,
3
,
Zhan
g
Jin
2
,
YuXiaomin
g
1
,
LiuYue
1
,
andChen
g
Xue
q
i
4
1
(
DataIntelli
g
enceS
y
stem ResearchCenter
,
Instituteo
f
Com
p
utin
g
Technolo
gy
,
ChineseAcadem
y
o
f
Sciences
,
Bei
j
in
g
100190
)
2
(
CAS Ke
y
Laborator
y
o
f
NetworkDataScienceand Technolo
gy
(
Instituteo
f
Com
p
utin
g
Technolo
gy
,
ChineseAcadem
y
o
f
Sciences
),
Bei
j
in
g
100190
)
3
(
Universit
y
o
f
ChineseAcadem
y
o
f
Sciences
,
Bei
j
in
g
100049
)
4
(
Instituteo
f
NetworkTechnolo
gy
∕
ICT
(
YANTAI
),
ChineseAcadem
y
o
f
Sciences
,
Yantai
,
Shandon
g
264005
)
Abstract Withthera
p
iddevelo
p
mentofinformationtechnolo
gy
,
Internethasbecomethe main
carrierof
p
ublico
p
inions
p
readin
g
.Allkindsof
p
ublico
p
inioneventscomeoutoneaftertheother
,
whichcanbe
q
uickl
y
s
p
readondifferentmediainashorttimeandreceiveextensiveattentionand
p
artici
p
ationfromlar
g
eGscaleInternetusers.Itma
y
alsotri
gg
erastron
g
reactionintheInternets
p
ace
andtherealsociet
y
,
andeveninducelar
g
eGscale massincidents.Therefore
,
q
uick monitorin
g
and
effectiveearl
y
warnin
g
ofonline
p
ublico
p
inioneventsbecomemoreandmoreim
p
ortant.Inres
p
onse
tothe
g
rowin
g
scaleofInternetinformationandtheex
p
andin
g
sco
p
eofitsdissemination
,
howto
discoveronline
p
ublico
p
inioneventsand
p
ushthe
p
recise
p
ersonalized monitorin
g
informationhas
becomethefocusofcurrent
p
ublico
p
iniona
pp
lications.Aimin
g
atthe
p
roblemthattheinterestis
hardtoca
p
tureincaseofthes
p
arsit
y
ofuserinteraction
,
ato
p
icrecommendation modelbasedon
HKN
(
hierarchicalknowled
g
enetwork
)
is
p
ro
p
osed.Themodelex
p
andsthesemanticsandincreases
p
otentialinformation association betweento
p
ics b
y
usin
g
hierarchicalknowled
g
e.It modelsthe
hierarchical knowled
g
e
,
to
p
ics
,
and users to obtain corres
p
ondin
g
embeddin
g
s. With those
embeddin
g
s
,
weuseamultiGla
y
er
p
erce
p
tronmatchin
g
modelto
p
redicttheCTR
(
clickthrou
g
hrate
)
.
Ex
p
erimentalresultsshowthatthe HKN modelout
p
erformsmulti
p
lebaselineal
g
orithmsb
y
6.7%
and4.9% ontheavera
g
eof
F
1
(
thebalanced
F
score
)
andAUC
(
theareaundercurve
)
metricsCTR
valueres
p
ectivel
y
.
评论