diff --git a/paper/Muhamad_Sulhan_-_Detecting_vandalism_on_English_Wikipedia_using_CRF.tex b/paper/23513014_Muhamad_Sulhan_-_Paper.tex
similarity index 100%
rename from paper/Muhamad_Sulhan_-_Detecting_vandalism_on_English_Wikipedia_using_CRF.tex
rename to paper/23513014_Muhamad_Sulhan_-_Paper.tex
diff --git a/paper/diagram_process.tex b/paper/diagram_process.tex
index ab2b258..990d530 100644
--- a/paper/diagram_process.tex
+++ b/paper/diagram_process.tex
@@ -1,7 +1,7 @@
 \begin{figure}[tp!]
 \centering
 \resizebox{0.3\textwidth}{!} {
-\tikzsetnextfilename{diagramprocess}
+\mytikzinput{diagramprocess}
 \begin{tikzpicture}[
 	framed,
 	nodes = {
diff --git a/paper/introduction.tex b/paper/introduction.tex
index 593b360..a857fe8 100644
--- a/paper/introduction.tex
+++ b/paper/introduction.tex
@@ -24,17 +24,15 @@
 assume that class distribution is balanced, while in real world cases this
 rarely happened.
 
-RF has the disadvantage in their the computation time especially when training
-the classification model.
-For a large dataset with more than 10,000 samples (like the PAN-WVC-10 cases)
-this could lead to hours of training time.
-One of the solution is by using Cascaded Random Forest (CRF) framework proposed
-by Bauman et al.
-\cite{baumann2013cascaded}.
-Their paper state that CRF give a fast training model time and increased
-performance compared to RF.
+Random Forest (RF) has the disadvantages in their the computation time
+especially when training the classification model.  For a large dataset with
+more than 10,000 samples (like the PAN-WVC-10 cases) this could lead to hours
+of training time.  One of the solution is by using Cascaded Random Forest (CRF)
+framework proposed by Bauman et al.  \cite{baumann2013cascaded}.  Their paper
+state that CRF give a fast training model time and increased performance
+compared to RF.
 
-This paper attempt to overcome the dataset imbalance problem on PAN-WVC-10 by
+This paper attempts to overcome the dataset imbalance problem on PAN-WVC-10 by
 applying resample and classifier technique that has never been used before on
 the dataset.
 The PAN-WVC-10 dataset is resampled using Local Neighborhood SMOTE (LNSMOTE)
diff --git a/paper/latexmk.sh b/paper/latexmk.sh
index 657936d..6a86dfa 100755
--- a/paper/latexmk.sh
+++ b/paper/latexmk.sh
@@ -1,3 +1,3 @@
 #!/bin/sh
 
-latexmk -pvc -view=pdf Muhamad_Sulhan_-_Detecting_vandalism_on_English_Wikipedia_using_CRF.tex
+latexmk -pvc -view=pdf 23513014_Muhamad_Sulhan_-_Paper.tex
diff --git a/paper/preamble.tex b/paper/preamble.tex
index 0cd818a..55edf9b 100644
--- a/paper/preamble.tex
+++ b/paper/preamble.tex
@@ -47,6 +47,13 @@
 \usetikzlibrary{backgrounds, shapes.geometric, positioning, patterns, external}
 	\tikzexternalize
 
+%% Make tikz generate PDF file to .tmp directory
+\makeatletter
+	\newcommand{\mytikzinput}[1]{%
+		\tikzsetnextfilename{tmp/#1}%
+	}
+\makeatother
+
 %% Pgfplots
 \usepackage{pgfplots}