Skip to content
GitLab
Menu
Projects
Groups
Snippets
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Sign in
Toggle navigation
Menu
Open sidebar
Marc Feger
Argument Relevance Presentation
Commits
bd0b948e
Commit
bd0b948e
authored
Sep 04, 2020
by
Marc Feger
Browse files
Refactor Methods
parent
8b8d9709
Pipeline
#45904
passed with stage
in 21 seconds
Changes
4
Pipelines
1
Hide whitespace changes
Inline
Side-by-side
bilder/BestPageRank.png
0 → 100644
View file @
bd0b948e
400 KB
bilder/Vectors.png
0 → 100644
View file @
bd0b948e
8.98 KB
bilder/Wordnet.png
0 → 100644
View file @
bd0b948e
30.6 KB
slides/methods.tex
View file @
bd0b948e
...
...
@@ -4,64 +4,50 @@
\begin{frame}
\frametitle
{
PageRank
}
\begin{columns}
\column
{
0.
6
\textwidth
}
\column
{
0.
5
\textwidth
}
\begin{itemize}
\item
PageRank by
\cite
{
PageRank1999
}
was originally used to evaluate relevant websites via their links
\item
Websites can be replaced by arguments
\item
Linking results due to the reuse of conclusions and premises
\item
In this work it was used:
\begin{itemize}
\item
Custom-made PageRank
\item
NetworkX PageRank
\item
NetworkX-Scipy PageRank
\end{itemize}
\item
PageRank originally used for websites
\item
Websites replaced by arguments
\end{itemize}
\begin{block}
{
Custom-made PageRank
}
\begin{equation*}
p
_
t(c
_
i) =
\left\{
\begin{array}
{
lr
}
(1 -
\alpha
)
\frac
{
1
}{
|D|
}
+
\alpha
\sum
_{
j
}{
\frac
{
p
_{
t-1
}
(c
_
j)
}{
|P'
_
j|
}
}
&
: t > 0
\\
\frac
{
1
}{
|D|
}
&
: t = 0
(1 -
\alpha
)
G
_{
rel
}
+
\alpha
L
_{
rel
}
&
: t > 0
\\
G
_{
rel
}
&
: t = 0
\end{array}
\right
.
\end{equation*}
\end{block}
\column
{
0.
4
\textwidth
}
\column
{
0.
5
\textwidth
}
\includegraphics
[scale=0.25]
{
bilder/ExampleGraph.png
}
\end{columns}
\end{frame}
\begin{frame}
\frametitle
{
WordNet
}
\begin{itemize}
\item
As knowledge based method
$
Sim
(
T
_
1
, T
_
2
)
$
of
\cite
{
Mihalcea2006similarity
}
was used
\item
$
Sim
(
T
_
1
, T
_
2
)
$
determines the semantic similarity of
$
T
_
1
,T
_
2
$
by mutually picking up highly similar concepts
\item
The concepts were determined via WordNet
\item
For
$
maxSim
$
the
$
CoSim
$
of
\cite
{
Wu1994distance
}
was used
\end{itemize}
\begin{block}
{$
Sim
(
T
_
1
, T
_
2
)
$}
\begin{equation*}
\frac
{
1
}{
2
}
(
\frac
{
\sum
_{
w
\in
{
T
_
1
}}{
maxSim(w, T
_
2)
\cdot
idf(w)
}}{
\sum
_{
w
\in
{
T
_
1
}}{
idf(w)
}}
+
\frac
{
\sum
_{
w
\in
{
T
_
2
}}{
maxSim(w, T
_
1)
\cdot
idf(w)
}}{
\sum
_{
w
\in
{
T
_
2
}}{
idf(w)
}}
)
\end{equation*}
\end{block}
\begin{itemize}
\item
Analogously, the average conceptual similarity between
$
T
_
1
$
and
$
T
_
2
$
was used as a weakened variant
\end{itemize}
\begin{columns}
\column
{
0.5
\textwidth
}
\begin{itemize}
\item
Knowledge-based method
\item
Conceptual similarity between conclusion and premise
\end{itemize}
\column
{
0.5
\textwidth
}
\includegraphics
[scale=0.33]
{
bilder/Wordnet.png
}
\end{columns}
\end{frame}
\begin{frame}
\frametitle
{
Similarity
}
\begin{itemize}
\item
For ranking the arguments, we measured the semantic similarity
between the premises and conclusions
\item
Argument were embedded word-wise in an averaged vector space
\item
The resulting similarity was calculated by using
$
Cos
(
c, p
)
$
\end{itemize}
\begin{block}
{
Embeddings used
}
BERT by
\cite
{
devlin2018bert
}
ELMo by
\cite
{
Peters:2018ELMo
}
GloVe by
\cite
{
pennington2014glove
}
\end{block}
\begin{columns}
\column
{
0.5
\textwidth
}
\begin{itemize}
\item
Semantic similarity
\item
Different embeddings
\item
BERT, ELMo and GloVe
\item
Similarity over
$
Cos
(
C, P
)
$
\end{itemize}
\column
{
0.5
\textwidth
}
\includegraphics
[scale=0.4]
{
bilder/Vectors.png
}
\end{columns}
\end{frame}
\begin{frame}
\frametitle
{
Sentiment
}
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment