Skip to content
GitLab
Commit 68fff80e authored by Marc Feger's avatar Marc Feger
Browse files

Add new presentation with vocabulary extraction

parent 26e9d4d5
Expand all Hide whitespace changes
Inline Side-by-side
bilder/Pipeline_Vocab.svg 0 → 100644
View file @ 68fff80e
This diff is collapsed.
master.tex
View file @ 68fff80e
......@@ -187,113 +187,54 @@ Department for Computer-Networks and Communication-Systems}
\end{itemize}
\end{frame}
\begin{frame}
\begin{center}
\includegraphics[width=0.8\textwidth]{bilder/Overview.png}
Overview by \cite{SchaeferStede2021}.
CA = Corpus Annotation,
AD = Argument Detection,
SD = Stance Detection,
CD = Claim Detection.
\end{center}
\end{frame}
\begin{frame}
\begin{center}
\includegraphics[width=0.75\textwidth]{bilder/AnnotationSchema.png}
Overview by \cite{SchaeferStede2021}.
\end{center}
\end{frame}
\begin{frame}
\begin{center}
\begin{columns}
\column{0.5\textwidth}
\begin{center}
\includegraphics[width=0.8\textwidth]{bilder/Corpus.png}
Corpora + IAA reported \cite{SchaeferStede2021}.
ADU = Argument Discourse Unit,
cx = Cohens Kappa
\end{center}
\column{0.5\textwidth}
\begin{center}
\includegraphics[width=0.8\textwidth]{bilder/PrevResults.png}
Results reported by \cite{SchaeferStede2021}.
ADU = Argument Discourse Unit,
cx = Cohens Kappa
\end{center}
\end{columns}
\includegraphics[scale=0.3]{bilder/Pipeline}
\end{center}
\end{frame}
\section{Current}
\begin{frame}
\begin{center}
\includegraphics[scale=0.3]{bilder/Pipeline}
\end{center}
\end{frame}
\begin{frame}{\#abortion}
\begin{center}
\begin{columns}
\column{0.5\textwidth}
\begin{center}
\includegraphics[scale=0.2]{bilder/Abortion_WordCloud.png}
\end{center}
\column{0.5\textwidth}
\begin{center}
\begin{itemize}
\item Topic: Abortion
\item Time: 2021-08-15 - 2021-10-15
\item Root-Tweets: 15.273
\begin{itemize}
\item Conversation-Starter: 2.250
\end{itemize}
\item Conversation-Tweets: 14.569
\end{itemize}
\end{center}
\end{columns}
\includegraphics[scale=0.3]{bilder/Pipeline_Vocab}
\end{center}
\end{frame}
\begin{frame}{\#abortion}
\begin{center}
\includegraphics[scale=0.3]{bilder/Abortion_Time.png}
\end{center}
\end{frame}
\begin{frame}{\#brexit}
\begin{frame}{Kialo}
\begin{center}
\begin{columns}
\column{0.5\textwidth}
\begin{center}
\includegraphics[scale=0.2]{bilder/Brexit_WordCloud.png}
\end{center}
\includegraphics[scale=0.1]{bilder/kialo.jpg}
\column{0.5\textwidth}
\begin{center}
\begin{itemize}
\item Topic: Brexit
\item Time: 2020-01-01 - 2020-03-01
\item Root-Tweets: 151.090
\begin{itemize}
\item Conversation-Starter: 24.806
\end{itemize}
\item Conversation-Tweets: 275.968
\end{itemize}
\end{center}
\end{columns}
\begin{itemize}
\item Labeled Arguments (Pro/Con)
\item Strict Guidelines
\item Reviewed
\end{itemize}
\end{columns}
\end{center}
\begin{tcolorbox}[colback=cyan!5!white,colframe=cyan!75!black,title=Kialo Claim]
A \textbf{claim} is, in its most basic form, a \textbf{single piece} of an \textbf{argument} made in a discussion.
[...]
This \textbf{allows} you to \textbf{inspect} each step of the \textbf{thinking behind} an \textbf{argument}.
\end{tcolorbox}
\end{frame}
\begin{frame}{\#brexit}
\begin{frame}{Datasets}
\begin{center}
\includegraphics[scale=0.3]{bilder/Brexit_Time.png}
\begin{itemize}
\item [Kialo]:
\begin{itemize}
\item[abortion]: 3.201 Claims
\item[brexit]: 2.554 Claims
\end{itemize}
\item [Twitter]:
\begin{itemize}
\item[abortion]: 2.250 Root-Tweets
\item[brexit]: 24.806 Root-Tweets
\end{itemize}
\end{itemize}
\end{center}
\end{frame}
......@@ -307,9 +248,9 @@ Department for Computer-Networks and Communication-Systems}
\begin{tabular}{@{}lll@{}}
\toprule
\textbf{Open} & \textbf{Closed} & \textbf{Other} \\ \midrule
ADJ & \textbf{AUX} & PUNCT \\
ADV & \textbf{CCONJ} & SYM \\
INTJ & DET & X \\
\textbf{ADJ} & \textbf{AUX} & PUNCT \\
\textbf{ADV} & \textbf{CCONJ} & SYM \\
INTJ & \textbf{DET} & X \\
NOUN & NUM & \\
PROPN & PART & \\
\textbf{VERB} & PRON & \\
......@@ -320,138 +261,141 @@ Department for Computer-Networks and Communication-Systems}
\end{center}
\column{0.6\textwidth}
\begin{itemize}
\item Universal POS Tagging
\item spaCy: en\_core\_web\_sm
\item Open v. Closed $\approx$ Content v. Function
\item Selection supposed by \cite{Knott1993}.
\item \textbf{ADJ}: \\
modify, specify nouns
\item \textbf{ADV}: \\
modify, specify verbs
\item \textbf{*VERB}: \\
typically signal events and actions
\item \textbf{*AUX}: \\
adds funct. or gramma. meaning
\item \textbf{*CCONJ}: \\
expresses a semantic relationship
\item \textbf{DET}: \\
express references in context
\item \textbf{*SCONJ}:\\
making one construction part of the other
\end{itemize}
\end{columns}
\end{center}
\end{frame}
\begin{frame}{POS-Tagging}
\begin{frame}{POS-Tag Distribution by words}
\begin{center}
\begin{columns}
\column{0.4\textwidth}
\begin{center}
\begin{table}[]
\resizebox{0.8\textwidth}{!}{%
\begin{tabular}{@{}lll@{}}
\toprule
\textbf{Open} & \textbf{Closed} & \textbf{Other} \\ \midrule
ADJ & \textbf{AUX} & PUNCT \\
ADV & \textbf{CCONJ} & SYM \\
INTJ & DET & X \\
NOUN & NUM & \\
PROPN & PART & \\
\textbf{VERB} & PRON & \\
& \textbf{SCONJ} & \\ \bottomrule
\end{tabular}%
}
\end{table}
\end{center}
\column{0.6\textwidth}
\begin{itemize}
\item \textbf{VERB}: \\
typically signal events and actions
\item \textbf{AUX}: \\
adds funct. or gramma. meaning
\item \textbf{CCONJ}: \\
links words or larger constituents,\\
expresses a semantic relationship
\item \textbf{SCONJ}:\\
making one construction part of the other,\\
join independent and a dependent clause
\end{itemize}
\end{columns}
\includegraphics[width=\textwidth]{bilder/PosDis.png}
\end{center}
\end{frame}
\begin{frame}{POS-Tagging Conversation Starter}
\begin{frame}{Vocabulary Selection}
\begin{itemize}
\item [1] Sorting Word-Lemma per POS-Tag by occurance
\item [2] Select Index with best Spearman-Rank-Correlation
\item [3] Intersection of all Dataset per POS-Tag till best Index
\end{itemize}
\end{frame}
\begin{frame}{Step 1: Example SCONJ}
\begin{center}
\includegraphics[scale=0.3]{bilder/Abortion_POS.png}
\begin{tabular}{lllll}
\toprule
{} & \textbf{abortion\_kialo} & \textbf{brexit\_kialo} & \textbf{abortion} & \textbf{brexit} \\
\midrule
0 & that & that & that & that \\
1 & if & if & if & if \\
2 & \textbf{because} & \textbf{as} & how & how \\
3 & when & \textbf{because} & when & \textbf{as} \\
4 & \textbf{as} & when & why & when \\
5 & where & since & \textbf{because} & why \\
6 & whether & for & \textbf{as} & \textbf{because} \\
7 & for & how & where & where \\
8 & while & while & while & for \\
9 & since & than & since & since \\
10 & how & where & for & after \\
11 & why & whether & until & while \\
12 & than & why & before & so \\
13 & upon & before & after & like \\
14 & after & after & whether & than \\
\bottomrule
\end{tabular}
\end{center}
\end{frame}
\begin{frame}{POS-Tagging Conversation Starter}
\begin{frame}{Step 2: Best Indices}
\begin{center}
\includegraphics[scale=0.3]{bilder/Brexit_POS.png}
\begin{tabular}{lll}
\toprule
\textbf{Tag} & \textbf{Spearman} & \textbf{Index} \\
\midrule
SCONJ & 0.93 & 26 \\
CCONJ & 0.83 & 10 \\
AUX & 0.91 & 26 \\
VERB & 1.0 & 10 \\
DET & 0.97 & 27 \\
ADJ & 1.0 & 10 \\
ADV & 0.89 & 30 \\
\bottomrule
\end{tabular}
\end{center}
\end{frame}
\begin{frame}{POS-Tagging Words (PWords)}
\begin{table}[]
\centering
\resizebox{\textwidth}{!}{%
\begin{tabular}{@{}llllllll@{}}
\toprule
\multicolumn{2}{c}{\textbf{SCONJ}} & \multicolumn{2}{c}{\textbf{CONJ}} & \multicolumn{2}{c}{\textbf{AUX}} & \multicolumn{2}{c}{\textbf{VERB}} \\ \midrule
\textit{\textbf{Brexit}} & \multicolumn{1}{l|}{\textit{\textbf{Abortion}}} & \textit{\textbf{Brexit}} & \multicolumn{1}{l|}{\textit{\textbf{Abortion}}} & \textit{\textbf{Brexit}} & \multicolumn{1}{l|}{\textit{\textbf{Abortion}}} & \textit{\textbf{Brexit}} & \textit{\textbf{Abortion}} \\ \midrule
that & \multicolumn{1}{l|}{that} & and & \multicolumn{1}{l|}{and} & is & \multicolumn{1}{l|}{is} & have & have \\
if & \multicolumn{1}{l|}{if} & but & \multicolumn{1}{l|}{but} & will & \multicolumn{1}{l|}{are} & get & get \\
how & \multicolumn{1}{l|}{how} & or & \multicolumn{1}{l|}{or} & be & \multicolumn{1}{l|}{be} & going & do \\
as & \multicolumn{1}{l|}{when} & + & \multicolumn{1}{l|}{+} & are & \multicolumn{1}{l|}{will} & know & want \\
when & \multicolumn{1}{l|}{why} & plus & \multicolumn{1}{l|}{yet} & have & \multicolumn{1}{l|}{can} & \textbf{leave} & know \\
why & \multicolumn{1}{l|}{because} & so & \multicolumn{1}{l|}{nor} & can & \multicolumn{1}{l|}{do} & \textbf{voted} & make \\
because & \multicolumn{1}{l|}{as} & both & \multicolumn{1}{l|}{both} & has & \multicolumn{1}{l|}{should} & want & \textbf{need} \\
where & \multicolumn{1}{l|}{where} & yet & \multicolumn{1}{l|}{neither} & was & \multicolumn{1}{l|}{was} & \textbf{see} & think \\
since & \multicolumn{1}{l|}{while} & either & \multicolumn{1}{l|}{so} & do & \multicolumn{1}{l|}{have} & think & let \\
for & \multicolumn{1}{l|}{since} & nor & \multicolumn{1}{l|}{either} & am & \multicolumn{1}{l|}{would} & let & \textbf{has} \\
after & \multicolumn{1}{l|}{after} & n & \multicolumn{1}{l|}{n} & would & \multicolumn{1}{l|}{has} & \textbf{done} & \textbf{say} \\
while & \multicolumn{1}{l|}{\textbf{until}} & • & \multicolumn{1}{l|}{plus} & \textbf{been} & \multicolumn{1}{l|}{am} & do & going \\
\textbf{so} & \multicolumn{1}{l|}{for} & neither & \multicolumn{1}{l|}{•} & were & \multicolumn{1}{l|}{\textbf{does}} & \textbf{go} & \textbf{had} \\
\textbf{like} & \multicolumn{1}{l|}{\textbf{before}} & \textbf{minus} & \multicolumn{1}{l|}{} & \textbf{could} & \multicolumn{1}{l|}{\textbf{being}} & \textbf{leaving} & \textbf{support} \\
\textbf{despite} & \multicolumn{1}{l|}{\textbf{whether}} & \textbf{daythe} & \multicolumn{1}{l|}{} & should & \multicolumn{1}{l|}{were} & make & \textbf{said} \\ \bottomrule
\end{tabular}%
}
\end{table}
\begin{frame}{Step 3: Vocabulary Selection}
\begin{center}
\resizebox{0.75\textwidth}{!}{
\begin{tabular}{llllllll}
\toprule
{} & \textbf{SCONJ} & \textbf{CCONJ} & \textbf{AUX} & \textbf{*VERB} & \textbf{DET} & \textbf{*ADJ} & \textbf{*ADV} \\
\midrule
0 & after & and & \textbf{be} & \textbf{be} & a & more & also \\
1 & as & both & can & \textbf{do} & all & other & always \\
2 & because & but & could & \textbf{have} & an & & as \\
3 & before & either & \textbf{do} & make & another & & even \\
4 & despite & nor & get & & any & & just \\
5 & for & or & \textbf{have} & & both & & long \\
6 & how & so & having & & each & & more \\
7 & if & yet & may & & either & & most \\
8 & once & & might & & every & & never \\
9 & since & & must & & half & & only \\
10 & so & & need & & no & & so \\
11 & than & & shall & & quite & & still \\
12 & that & & should & & some & & then \\
13 & though & & to & & such & & too \\
14 & unless & & will & & that & & very \\
15 & until & & would & & the & & well \\
16 & upon & & & & these & & \\
17 & when & & & & this & & \\
18 & where & & & & those & & \\
19 & whether & & & & what & & \\
20 & while & & & & whatever & & \\
21 & why & & & & which & & \\
22 & with & & & & whose & & \\
\bottomrule
\end{tabular}
}
\end{center}
\end{frame}
\begin{frame}{Pre-Selection}
\begin{frame}{First Impression}
Using vocab. with SCONJ, CCONJ, AUX, DET:
\begin{itemize}
\item Conditions:
\begin{itemize}
\item [1] At least one word in PWords
\item [2] At least 200 characters long
\end{itemize}
\item \#Abortion: $\Rightarrow$ 1.400 Candidates
\item \#Brexit: $\Rightarrow$ 11.842 Candidates
\item [$\Rightarrow$] 0.069\% dropoff in abortion
\item [$\Rightarrow$] 0.066\% dropoff in brexit
\item [$\Rightarrow$] Dropoff = Adds, Allegations, Comments, Links etc.
\item [$\Rightarrow$] Root-Tweets $\geq$200 chars) could be better
\end{itemize}
\end{frame}
\begin{frame}
\begin{tcolorbox}[colback=cyan!5!white,colframe=cyan!75!black,title=Example 1: \#Brexit]
\textbf{How} ironic \textbf{is} this? \textbf{For} 4 years, \textcolor{orange}{\#Remainers} predicted \textbf{that} \textcolor{orange}{\#Brexit} \textbf{would} cause a catastrophic national crisis. Now \textbf{that} Brexit \textbf{has} happened at long last, we \textbf{are} indeed facing a catastrophic national crisis, \textbf{but} it \textbf{has} absolutely nothing to \textbf{do} with Brexit. Go figure \textbf{that} one out.
%\textcolor{red}{RT} \textcolor{magenta}{@SaysSheToday}: The $[$Dixie Chicks$]_{c}$ were attacked just for $[$using 1A right$]_{p_1}$ to say they were ashamed of GWB. They $[$didn’t commit treason$]_{c}$ $[$like the \textcolor{orange}{\#47Senators}$]_{p_2}$
\end{tcolorbox}
\end{frame}
\begin{frame}
\begin{tcolorbox}[colback=cyan!5!white,colframe=cyan!75!black,title=Example 2: \#Brexit]
Labour cannot win the next general election with \textcolor{magenta}{@Keir\_Starmer} \textbf{because} voters \textbf{do}n’t trust him. Starmer \textbf{is} a proven liar \textbf{and} fraud who tried to cheat 17.4m voters out of their \textcolor{orange}{\#Brexit} referendum victory. Starmer \textbf{is} unelectable. \textcolor{orange}{\#LabourDebate} \textcolor{orange}{\#labourhustings}
%\textcolor{red}{RT} \textcolor{magenta}{@SaysSheToday}: The $[$Dixie Chicks$]_{c}$ were attacked just for $[$using 1A right$]_{p_1}$ to say they were ashamed of GWB. They $[$didn’t commit treason$]_{c}$ $[$like the \textcolor{orange}{\#47Senators}$]_{p_2}$
\end{tcolorbox}
\end{frame}
\begin{frame}
\begin{tcolorbox}[colback=cyan!5!white,colframe=cyan!75!black,title=Example 3: \#Brexit]
\textcolor{orange}{\#SaturdayThought} \textbf{If} \textcolor{orange}{\#PMs} "oven-ready Brexit deal" \textbf{is so} "oven-ready," \textbf{why has} he not got full agreement of the EU \textbf{yet}? \textbf{After} all, \textcolor{magenta}{@Conservatives} keep telling us \textbf{that} \textcolor{orange}{\#Brexit} \textbf{is} over now.\textcolor{orange}{\#JustAsking}
%\textcolor{red}{RT} \textcolor{magenta}{@SaysSheToday}: The $[$Dixie Chicks$]_{c}$ were attacked just for $[$using 1A right$]_{p_1}$ to say they were ashamed of GWB. They $[$didn’t commit treason$]_{c}$ $[$like the \textcolor{orange}{\#47Senators}$]_{p_2}$
\end{tcolorbox}
\end{frame}
\begin{frame}
\begin{tcolorbox}[colback=cyan!5!white,colframe=cyan!75!black,title=Example 4: \#Abortion]
POV: Stop saying NOBODY \textbf{is} “pro-abortion." It\textbf{'s} \textcolor{orange}{\#stigmatising}. \textbf{Its} simply inaccurate, Plenty of folk \textbf{are} unapologetically \textcolor{orange}{\#proabortion}; \textbf{because} there\textbf{'s} nothing wrong with \textcolor{orange}{\#Abortion} -it\textbf{`s} a necessary health care procedure \textbf{that} saves lives.
%\textcolor{red}{RT} \textcolor{magenta}{@SaysSheToday}: The $[$Dixie Chicks$]_{c}$ were attacked just for $[$using 1A right$]_{p_1}$ to say they were ashamed of GWB. They $[$didn’t commit treason$]_{c}$ $[$like the \textcolor{orange}{\#47Senators}$]_{p_2}$
\end{tcolorbox}
\end{frame}
\section{Next}
\begin{frame}
\begin{itemize}
\item [1] More controverse topics
\item [2] Cleaning of PWords
\item [3] PWords and POS-Distribution for Kialo
\item [4] Annotation / Quality-Check on Candidate-Subset
\item [1] Selection on SCONJ, CCONJ, AUX
\item [2] Selection sampling:
\begin{itemize}
\item [] \textbf{Sample 1}: Dropoff
\item [] \textbf{Sample 2}: Root-Tweets with length $<$ 200
\item [] \textbf{Sample 3}: Root-Tweets with length $\geq$200
\end{itemize}
\item [3] Annotation of samples
\item [4] Fine-Tuning of argumentive vocabulary
\end{itemize}
\end{frame}
......
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment