Predicting Document Creation Times in News Citation Networks Andreas - - PowerPoint PPT Presentation
Predicting Document Creation Times in News Citation Networks Andreas Spitz 1 , Jannik Strtgen 2 , and Michael Gertz 1 April 23, 2018 TempWeb 2018, Lyon 1 Database Systems Research Group 2 Bosch Center for Artificial Intelligence Heidelberg
1 Database Systems Research Group 2 Bosch Center for Artificial Intelligence
Heidelberg University, Germany Germany
1
2
◮ Politics and business feeds ◮ 34 English news outlets
◮ 2 years (Nov 2015 - Oct 2017) ◮ 244.6 thousand articles ◮ 367.2 thousand edges
◮ Hyperlinks in the article body ◮ Publication dates ◮ Temporal expressions 3
4
2016−01 2016−07 2017−01 2017−07 2016−01 2016−07 2017−01 2017−07 20 40 60 5 10 15 1 2 3 0.0 0.2 0.4 0.6
5
6
7
◮ Citation network topology ◮ Publication dates of adjacent articles ◮ Temporal expressions in adjacent articles 8
◮ Citation network topology ◮ Publication dates of adjacent articles ◮ Temporal expressions in adjacent articles ◮ Not the metadata of the article itself ◮ Not the article content 8
◮ Incoming degree ◮ Outgoing degree ◮ Undirected degree 9
◮ Incoming degree ◮ Outgoing degree ◮ Undirected degree
◮ Betweenness centrality ◮ Incoming closeness centrality ◮ Outgoing closeness centrality ◮ Page Rank centrality 9
◮ Incoming degree ◮ Outgoing degree ◮ Undirected degree
◮ Undirected local clustering coefficient
◮ Betweenness centrality ◮ Incoming closeness centrality ◮ Outgoing closeness centrality ◮ Page Rank centrality 9
10
◮ good with publication dates of
◮ bad with publication dates of
11
◮ good with publication dates of
◮ bad with publication dates of
11
◮ 30.8% of feature values are missing ◮ 89.6% of articles are missing at least one feature 12
◮ 30.8% of feature values are missing ◮ 89.6% of articles are missing at least one feature
◮ Column mean of the feature 12
◮ BASE: Baseline (average publication date of adjacent articles) ◮ LR: Linear regression ◮ BAY: Bayesian ridge regression (Laplace model) ◮ RF: Random forest ◮ GB: Gradient boosting (Laplace distribution, decision trees) ◮ SVM: Support vector machine (radial kernel) ◮ NN: Neural network (feedforward, one hidden layer) 13
14
in+out all in
BASE LR BAY NN RF GB SVM BASE LR BAY NN RF GB SVM 50 100 150 200 250 50 100 150 200 250
regression method absolute error (days)
method
BASE LR BAY NN RF GB SVM
15
in+out all in
20 40 60 20 40 60 25 50 75 100 25 50 75 100
absolute error (days) recall (percentage of predictions < absolute error)
method
BASE LR BAY NN RF GB SVM
16
max
T
min
T
in
µ
T
µ
T
in
min
T
max
T
in
max
X
in
µ
X
in
c
pr
σ
T
σ
X
in
c
cl,out
span
T
σ
T
in
span
T
in
min
X
in
span
X
in
c
cl,in
min
Dist
deg
µ
Dist
deg
in
deg
all
max
Dist
c
btw
cc σ
Dist
10−3 10−2 10−1 100
temporal expression temporal network
17
max
T
min
T
in
deg
µ
T
min
Dist
deg
in
c
pr
σ
T
σ
T
in
µ
T
in
deg
all
span
T
in
max
T
in
µ
X
in
min
T
µ
Dist
c
btw
max
X
in
max
Dist
span
X
in
σ
X
in
span
T
min
X
in
c
cl,out
σ
Dist
c
cl,in
cc 10−5 10−4 10−3 10−2 10−1 100
temporal expression temporal network
18
◮ Focus on anchored links inside the article body ◮ Constructed like a citation network between articles
◮ Can be framed as a regression problem ◮ Average prediction error of 3 weeks ◮ Temporal network features are most discriminative 19
◮ [data] News citation network (including URLs) ◮ [data] Temporal annotations ◮ [code] Publication date prediction
20
◮ [data] News citation network (including URLs) ◮ [data] Temporal annotations ◮ [code] Publication date prediction
20