Text Network Constructs

layout: true
class: center
---
class: middle
background-color: white

.title[Social Media Text Networks]     
.subtitle[How to do it and when you should: from trolls to journalists]  
   
   
.author[Laila A. Wahedi, PhD]  
.date[November 1, 2018]   
   
.institution[McCourt School of Public Policy, Georgetown University]

<img style="width:20%"  src="../figures/seal.png">
   
           
           
           
.center[]       
        
.footnote[
 Created with [{Remark.js}](http://remarkjs.com/) using
 [{Markdown}](https://daringfireball.net/projects/markdown/) +
 [{MathJax}](https://www.mathjax.org/) +
 [{Liminal}](http://www.jmlilly.net/liminal.zip)
 ]

---
layout: true
name: main
class: center
.footnote_left[ » Slides at https://wahedi.us under <i>Current Presentation</i> ]
.footnote_right[ Laila A. Wahedi • @lwahedi • law98@georgetown.edu ]

---
template: main
background-color: #F2DC5D
<div>
	<h1>  Motivation: Why Social Media Data?</h1>
	<div style="width:65%; margin: 0 auto">
		<img style="width:75%"  src="../figures/net_text/observe.png">
		<h2 style="color:black;">Observe Interactions</h2>
	</div>
</div>

---
template: main
background-color: #AF002B
<div>
	<h1 style='color:white'>Motivation: Why Social Media Data?</h1>
	<div style="width:50%; float:left;" >
		<img style="width:70%"  src="../figures/net_text/observe_types.png">
		<h2 style="color:white;">Not All The Same</h2>
	</div>
	<div style="width:50%; float:left;" >
		<img style="width:70%"  src="../figures/net_text/compare.png">
		<h2 style="color:white;">Relation ≠ Social</h2>
	</div>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Today: Text-Based Constructions</h2>
<img src="../figures/line.png" class="underline">
   
   
   
<div style="width:80%; margin: 0 auto;">
	<table>
		<tr>
			<td><img style="width:75%" src="../figures/net_text/observe_author.png"></td>
			<td><img style="width:75%" src="../figures/net_text/observe_bipartite.png"></td>
			<td><img style="width:75%" src="../figures/net_text/observe_content.png"></td>
		</tr>
		<tr>
			<td><h3 style="color:white;">Author Similarity</h3></td>
			<td><h3 style="color:white;">Topic Bipartite</h3></td>
			<td><h3 style="color:white;">Content Similarity</h3></td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<div style="width:100%;">
	<br>
	<br><br>
	<h2 style="color:white; text-align:left; margin-bottom:0px">Before Constructing a Network...</h2>
	<br><br> 
	<h2 style="color:white; text-align:right; margin-bottom:0px">What Do We Want From Relational Data?</h2>  
	<br><br> 
	<div style="width:50%; margin:0 auto">
		<img style="width:70%" src="../figures/net_text/connection.png">
	</div>
</div>

---
layout: true
name: analysis
template: main
background-color: #01BAEF

<div style="width:100%;">
	<h1 style='color:white'>Types of Analysis:</h1>
	<div style="width:95%; margin:0 auto;" >
		{{content}}
	</div>
</div>

---
<img style="width:75%" src="../figures/net_text/net_analysis.png">
---
<img style="width:100%" src="../figures/net_text/desc_structure.png">
---
<img style="width:100%" src="../figures/net_text/desc_cluster.png">
---
<img style="width:100%" src="../figures/net_text/desc_cent.png">
---
<img style="width:100%" src="../figures/net_text/flow.png">
---
<img style="width:100%" src="../figures/net_text/structure_flow.png">
---
<img style="width:100%" src="../figures/net_text/struct_behavior.png">
---
<img style="width:100%" src="../figures/net_text/clust_behavior.png">
---
<img style="width:100%" src="../figures/net_text/ties_behavior.png">

---
layout: false
template: main
background-color: #01BAEF
class: center

<h2 class="header">Social vs Other Relational Data:</h2>
<img src="../figures/line.png" class="underline">
   
   
   
<div style="width:100%; margin: 0 auto;">
	<table>
		<tr>
			<td><img style="width:90%" src="../figures/net_text/compare.png"></td>
			<td><img style="width:100%" src="../figures/net_text/differences.png"></td>
			<td><img style="width:80%" src="../figures/net_text/attribute_sim.png"></td>
			<td><img style="width:83%" src="../figures/net_text/attribute_dist.png"></td>
		</tr>
		<tr>
			<td><h3 style="font-size:1em;color:white;">Compare Attributes / Behaviors</h3></td>
			<td><h3 style="font-size:1em;color:white;">Effect of Similarity in Attribute</h3></td>
			<td><h3 style="font-size:1em;color:white;">Relationships between Attributes</h3></td>
			<td><h3 style="font-size:1em;color:white;">Distribution of Attribute Values</h3></td>
		</tr>
		<tr>
			<td colspan=4> </td>
		</tr>
		<tr>
			<td> </td>
			<td colspan=3 style="text-align:right;">
				<h3 style="color:white;">Same methods / metrics, <br>different meanings</h3>
			</td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity:  What?</h2>
<img src="../figures/line.png" class="underline">
   
   
   
<div style="width:100%; margin: 0 auto;">
	<table>
		<tr>
			<td><img style="width:70%" src="../figures/net_text/text_mat.png"></td>
			<td><img style="width:100%" src="../figures/net_text/protractor.png"></td>
			<td><img style="width:55%" src="../figures/net_text/connection.png"></td>
		</tr>
		<tr>
			<td><h3 style="font-size:1em;color:white;">1. Create Author Vectors</h3></td>
			<td><h3 style="font-size:1em;color:white;">2. Calculate Distance</h3></td>
			<td><h3 style="font-size:1em;color:white;">3. Weighted / Thresholded Network</h3></td>
		</tr>
		<tr>
			<td colspan=4> </td>
		</tr>
		<tr>
			<td> </td>
			<td colspan=3 style="text-align:right;">
				<h3 style="color:white;">Relate Authors By Rhetoric</h3>
			</td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: What?</h2>
<img src="../figures/line.png" class="underline">
   
   
<div style="width:100%;">
	<h3 style="text-align:left; color:white;margin-bottom:0.7em;"> Captures variation in how individuals speak</h3>
	<table>
		<tr>
			<td><div class="redbutton"><h3 style = "color:white; margin:.5em">Dialect / Region</h3></div></td>
			<td><div class="redbutton"><h3 style = "color:white; margin:.5em">Education / Profession</h3></div></td>
			<td><div class="redbutton"><h3 style = "color:white; margin:.5em">Ideological Grouping</h3></div></td>
		</tr>
		<tr>
			<td><div class="redbutton"><h3 style = "color:white; margin:.5em">Age</h3></div></td>
			<td><div class="redbutton"><h3 style = "color:white; margin:.5em">Topic</h3></div></td>
			<td><div class="redbutton"><h3 style = "color:white; margin:.5em">Subculture</h3></div></td>
		</tr>
	</table>
</div>
<div>
	<br><br>
</div>
<div style="width:100%;">
	<h3 style="text-align:left; color:white;margin-bottom:0.7em;"> Depends on Sample Variation</h3>
	<table>
		<tr>
			<td style="width:50%;">
				<div class="redbutton"><h3 style = "color:white; margin:.5em">Academic Twitter vs Random Sample</h3></div>
			</td>
			<td style="width:50%;">
				<div class="redbutton"><h3 style = "color:white; margin:.5em">#maga vs<br> #trump</h3></div>
			</td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: Why?</h2>
<img src="../figures/line.png" class="underline">
   
   
   
<div style="width:100%;">
	<table>
		<tr>
			<td><img style="width:80%" src="../figures/net_text/media_compare.png"></td>
			<td><img style="width:80%" src="../figures/net_text/hats.png"></td>
			<td><img style="width:80%" src="../figures/net_text/hats_grp.png"></td>
		</tr>
		<tr>
			<td class="narrow"><h3 style="font-size:1em;letter-spacing:0.01em;color:white;">Subcultures Exposed to Similar Content</h3></td>
			<td class="narrow"><h3 style="font-size:1em;letter-spacing:0.01em;color:white;">Capture Unobserved Similarities</h3></td>
			<td class="narrow"><h3 style="font-size:1em;letter-spacing:0.01em;color:white;">Relationships between Subcultures</h3></td>
		</tr>
		<tr>
			<td colspan=3> </td>
		</tr>
		<tr>
			<td colspan=3 style="text-align:right;"><h3 style="color:white;">Not independent,<br>Even if not social media friends </h3></td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: How? Author Vectors</h2>
<img src="../figures/line.png" class="underline">
   
  
   
   
<h3 style="text-align:left; color:white;"> Start with Word Embeddings</h3>
<div style="width:100%; float:left">
	<img style="width:75%" src="../figures/net_text/word2vec.gif">
	<p style="font-size:0.8em;color:white; ">From: 
		<a href="https://multithreaded.stitchfix.com/blog/2016/05/27/lda2vec/">
			https://multithreaded.stitchfix.com/blog/2016/05/27/lda2vec/
		</a>
	</p>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: How? Author Vectors</h2>
<img src="../figures/line.png" class="underline">
   
  
   
   
<h3 style="text-align:left; color:white;"> Add a Vector for Author</h3>
<div style="width:100%; float:left">
	<img style="width:75%"  src="../figures/net_text/doc2vec.png">
	<p style="font-size:0.8em;color:white; ">From: 
		<a href="https://medium.com/scaleabout/a-gentle-introduction-to-doc2vec-db3e8c0cce5e">
			https://medium.com/scaleabout/a-gentle-introduction-to-doc2vec-db3e8c0cce5e
		</a>
	</p>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: How? Author Vectors</h2>
<img src="../figures/line.png" class="underline">
   
<div style="width:100%;">
	<h3 style="text-align:left; color:white;margin-bottom:0.7em;"> Calculate Distance</h3>
	<table style="width:85%;">
		<tr>
			<td>
				<img style="width:60%; margin-bottom:0px;" src="../figures/net_text/cos_sim.png">
				<h3 style = "font-size:0.5em;color:white;margin-top:0px;margin-bottom:0px;">Credit: Wikipedia</h3>
			</td>
			<td style="width:15%;"></td>
			<td><img style="width:40%"  src="../figures/net_text/hats.png"></td>
		</tr>
		<tr>
			<td><h3 style = "font-size:1em;color:white;">Cosine Similarity</h3></td>
			<td style="width:15%;"></td>
			<td><h3 style = "font-size:1em;color:white;">Euclidean Distance<br>(Check Scale)</h3></td>
		</tr>
	</table>
</div>
<div style="width:100%;">
	<h3 style="text-align:left; color:white;margin-bottom:0.7em;"> Construct Network</h3>
	<table style="width:85%">
		<tr>
			<td><img style="width:60%"  src="../figures/net_text/weighted.png"></td>
			<td style="width:15%;"></td>
			<td><img style="width:40%"  src="../figures/net_text/threshold.png"></td>
		</tr>
		<tr>
			<td><h3 style = "font-size:1em;color:white;">Weighted</h3></td>
			<td style="width:15%;"></td>
			<td><h3 style = "font-size:1em;color:white;">Threshold</h3></td>
		</tr>
	</table>
</div>

---
template: main
background-color: #AF002B

<h2 class="header">Caution: Author Vectors</h2>
<img src="../figures/line.png" class="underline">
   
   
   
   
   
<div style="width:80%;margin:0 auto;">
	<table>
		<tr>
			<td style="padding-right: 25;"><img style="width:80%"  src="../figures/net_text/folder.png"></td>
			<td style="padding-left: 25;"><img style="width:80%"  src="../figures/net_text/paper.png"></td>
		</tr>
		<tr>
			<td style="padding-right: 25;"><h3 style = "font-size:1em;color:white;">Need Lots of Data</h3></td>
			<td style="padding-left: 25;"><h3 style = "font-size:1em;color:white;">Tweets Are Noisy</h3></td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: How? <br>N-Gram Comparison</h2>
<img src="../figures/line.png" class="underline">
   
   
<div style="width:50%; float:center">
	<img style="width:30%; margin-bottom:0px"  src="../figures/net_text/bag.png">
	<h4 style = "color:white; ">With a Bag of n-Grams...</h4>
</div>
   
<div style="width:100%;">
	<h3 style="text-align:left; color:white;margin-bottom:0.7em;"> Calculate Distance:</h3>
	<table>
		<tr style="padding-top: 0px">
			<td style="width:25%;vertical-align:middle;padding-left:0px;padding-right:7px;">
				<div class="redbutton">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Unique shared n-grams in support range (undirected)</h3>
				</div>
			</td>
			<td style="width:28%;vertical-align:middle;padding-left:7px;padding-right:7px;">
				<div class="redbutton">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Shared n-grams instances in support range (undirected)</h3>
				</div>
			</td>
			<td style="width:20%;vertical-align:middle;padding-left:7px;padding-right:7px;">
				<div class="redbutton">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Percentage of n-grams in common (directed)</h3>
				</div>
			</td>
			<td style="width:27%;vertical-align:middle;padding-left:7px;padding-right:0px;">
				<div class="redbutton">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Number / percent of posts with common n-gram</h3>
				</div>
			</td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: Insights</h2>
<img src="../figures/line.png" class="underline">
   
   
<div style="width:100%;float:center;">
<span style="width:50%;float:center;">
	<img style="width:16%; margin-bottom:0px; "  src="../figures/net_text/centralization.png">
	<h2 style = "color:white;">Centralization</h2>
</span>
</div>
   
   
<div style="width:100%;">
	<table>
		<tr style="padding-top: 0px">
			<td style="width:40%;vertical-align:middle;">
				<div class="redbutton" style="padding-top:10px;padding-bottom:10px;">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Uninformative Vectors.<br>Check Text Quantity/Quality</h3>
				</div>
			</td>
			<td style="width:40%;vertical-align:middle;">
				<div class="redbutton">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;"> <br>Core of Generalists<br> </h3>
				</div>
			</td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: Insights</h2>
<img src="../figures/line.png" class="underline">
   
   
<div style="width:100%;float:center;">
<span style="width:50%;float:center;">
	<img style="width:16%; margin-bottom:0px; "  src="../figures/net_text/clusters.png">
	<h2 style = "color:white;">Strong Clustering</h2>
</span>
</div>
   
   
<div style="width:100%;">
	<table>
		<tr style="padding-top: 0px">
			<td style="width:40%;vertical-align:middle;">
				<div class="redbutton" style="padding-top:10px;padding-bottom:10px;">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Segregate Based On Rhetoric. Possible Sub-cultures</h3>
				</div>
			</td>
			<td style="width:40%;vertical-align:middle;">
				<div class="redbutton" style="padding-top:10px;padding-bottom:10px;">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Do individuals follow or interact across sub-cultures?</h3>
				</div>
			</td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 class="header">Author Similarity: Insights</h2>
<img src="../figures/line.png" class="underline">
   
<div style="width:100%;float:center;">
	<span style="width:50%;float:center;">
		<img style="width:15%; margin-bottom:0px; "  src="../figures/net_text/redundant.png">
		<h2 style = "color:white;">Decentralized</h2>
	</span>
</div> 
<div style="width:100%;">
	<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;text-align:left;">High Clustering Coefficient</h3>
	<table>
		<tr style="padding-top: 0px">
			<td style="width:43%;vertical-align:middle;">
				<div class="redbutton" style="padding-top:10px;padding-bottom:10px;">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">High Similarity. Check that author vectors are informative. Sufficient data?</h3>
				</div>
			</td>
			<td style="width:43%;vertical-align:middle;">
				<div class="redbutton" style="padding-top:10px;padding-bottom:10px;">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">May be a homogenous, (or very heterogenous) population.</h3>
				</div>
			</td>
		</tr>
	</table>

<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;text-align:left;">Low Clustering Coefficient </h3>
	<table>
		<tr style="padding-top: 0px">
			<td style="width:40%;vertical-align:middle;">
				<div class="redbutton" style="padding-top:10px;padding-bottom:10px;">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">Individuals who speak similarly are not similar to others like them. Sufficient data?</h3>
				</div>
			</td>
			<td style="width:40%;vertical-align:middle;">
				<div class="redbutton" style="padding-top:10px;padding-bottom:10px;">
					<h3 style="font-size:1em;color:white;margin:.5em;letter-spacing:0.01em;">May be capturing topics, if population discusses random distribution of topics (Unlikely).</h3>
				</div>
			</td>
		</tr>
	</table>
</div>

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Author Similarity: Insights</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<center><div style="width:100%; ">
<div style="width:50%;">
<img style="width:30%; margin-bottom:0px; "  src="../figures/net_text/centrality.png">
<h2 style = "color:white;">Centrality </h2>
</div>
</div></center>
<div style="width:100%; margin-bottom:1em">
<h4 style = "color:white;margin-bottom:0px; text-align:left">Degree, Closeness</h4>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">High: more general user, or less informative vector</h4>
</div>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;margin-bottom:2.5em">Low: More distinct</h4>
</div>
</div>
<br>
<div style="width:100%;">
<h4 style = "color:white; text-align: left; margin-bottom:0px; margin-top:1em">Betweenness</h4>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">Fits in multiple sub-cultures</h4>
</div>
</div>

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Author Similarity: Questions</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">

<h3 style="text-transform: none; text-align:left; color:white;">As Dependent Variable</h3>

<div style="width:48%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">What nodal attributes push someone to the edge of a subculture?</h4>
</div>

<div style="width:48%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">Is there more homogeneity among the users of different hashtags?</h4>
</div>
<div style="width:48%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">What sort of individuals bridge communities?</h4>
</div>

<div style="width:100%; ">
<h2 style="text-align:left; margin-top:8em; color:white;">As Explanatory Variable</h2>
<div style="width:48%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">Is there a relationship between clusters and sub-culture labels?</h4>
</div>
<div style="width:48%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">How do the resulting clusters relate to one another?</h4>
</div>
</div>

---
template: main
class: middle
background-color: #78881f

---
template: main
background-color: #78881f

<div style="width:31%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#01BAEF; float:left">
<h4 style = "color:white; margin:.5em">Tweet text and time</h4>
</div>

<div style="width:31%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#01BAEF; float:left">
<h4 style = "color:white; margin:.5em">Url redirects</h4>
</div>
<div style="width:31%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#01BAEF; float:left">
<h4 style = "color:white; margin:.5em">Is quote or retweet</h4>
</div>
<br>
<div style="clear: both; float: left; display: block; position: relative; margin-top: 2em">
<h3 style="text-transform: none; text-align:left; color:white;display:inline-block;clear:both;">Don't have</h3>
</div>

<div style="clear: both; float: left; display: block; position: relative;width:100%">
<div style="width:31%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">Friends / Followers</h4>
</div>

<div style="width:31%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">Retweet user id</h4>
</div>
<div style="width:31%; border: 5px solid #FFFFFF; border-radius: 25px; background-color:#AF002B; float:left">
<h4 style = "color:white; margin:.5em">Interactions</h4>
</div>
</div>

---
template: main
background-color: #78881f

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Russian Trolls</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">
<br>
<h2 style="text-transform: none; text-align:left; color:white;">What can we learn about how the IRA operated?</h2>
<h3 style="text-transform: none; text-align:left; color:white;">Can we distinguish authors?</h3>
<h4 style="text-transform: none; text-align:right; color:white;"> If agents handle multiple accounts, <br>may observe similarities in phrasing and rhetoric</h4>
<h4 style="text-transform: none; text-align:right; color:white;"> Alternatively: Agents may share accounts or<br> develop unique personas for each.</h4>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<h2 style="text-transform: none; text-align:left; color:white;">How coordinated were the operations?</h2>
<h3 style="text-transform: none; text-align:left; color:white;">Can we distinguish classes of account with similar purposes?</h3>
<h4 style="text-transform: none; text-align:right; color:white;"> If accounts are created for specific purposes, <br>expect them to use similar language</h4>

---
template: main
background-color: #78881f

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Topic Bipartite: What? </h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">

<div style="width:100%">
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/topic.png">
<h3 style = "color:white; ">1. Topic Model</h3>
</div>

<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/bipartite_net.png">
<h3 style = "color:white; ">2. Ties from user to topic</h3>
</div>

<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/projection.png">
<h3 style = "color:white; ">3. Optional: Project to unipartite</h3>
</div>
</div>

<div style="width:100%; ">
<h2 style="text-align:right; margin-top:10em; color:white;">What makes authors similar? <br> How do topics relate?</h2>
</div>

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Topic Bipartite: Why?</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">

<div style="width:100%">
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/chatting.png">
<h4 style = "color:white; ">What are people talking about?</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/connection.png">
<h4 style = "color:white; ">How do people cluster<br> around topics?</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/translator.png">
<h4 style = "color:white; ">How do topics <br>cluster?</h4>
</div>
</div>

<div style="width:100%; ">
<h2 style="text-align:right; margin-top:10em; color:white;">Describe the structure of <br>the rhetorical space<br>
</h2>
</div>

---
template: main
background-color: #01BAEF

.left-column[
<center><div style="width:100%; ">
<div style="width:100%;">
<img style="width:30%; margin-bottom:0px; "  src="../figures/net_text/centralization.png">
<h2 style = "color:white;">Centralization</h2>
</div>
</div></center>
<br>
<div style="width:100%; ">

<div style="width:95%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">Some topics are universal, while others are more niche</h4>
</div>
</div>]

.right-column[
<center><div style="width:100%; ">
<div style="width:100%;">
<img style="width:30%; margin-bottom:0px; "  src="../figures/net_text/clusters.png">
<h2 style = "color:white;">Strong Clustering</h2>
</div>
</div></center>
<br>
<div style="width:100%; ">

<div style="width:95%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; text-align:center;">Insular communities sharing very different information. Polarization</h4>
</div>
<br>
<h2 style = "color:white;">Weak Clustering</h2>
</div>

<div style="width:95%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; text-align:center">Many topics discussed across whole network. Homogenous information environment</h4>
</div>
</div>]

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Topic Bipartite: Insights</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<center><div style="width:100%; ">
<div style="width:50%;">
<img style="width:30%; margin-bottom:0px; "  src="../figures/net_text/centrality.png">
<h2 style = "color:white;">Centrality </h2>
</div>
</div></center>
<div style="width:100%; margin-bottom:1em">
<h4 style = "color:white;margin-bottom:0px; text-align:left">Degree, Closeness</h4>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;margin-bottom:2.4em ">Bipartite: General topic, transcends insular communities</h4>
</div>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;">Projection: General topic, or related many others within a cluster</h4>
</div>
</div>
<div style=" float: left; width:100%">

<h4 style = "color:white; text-align: left; margin-bottom:0px; margin-top:1em">Betweenness</h4>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;margin-bottom:3.4em ">Bridges communities </h4>
</div>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">Possible common ground in distinct information environments </h4>
</div>
</div>

---
template: main
background-color: #78881f

<h2 style="text-transform: none; text-align:left; color:white;">How coordinated were the operations?</h2>
<h3 style="text-transform: none; text-align:left; color:white;">To what extent do the trolls communicate unified messages? </h3>
<h4 style="text-transform: none; text-align:right; color:white;">If well coordinated, we expect to see groups of trolls <br>promoting similar topics.</h4>

<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<h4 style="text-transform: none; text-align:left; color:white;">Note: topic modeling on Twitter data notoriously hard. Demonstrating with hashtags.</h4>

---
template: main
background-color: #78881f

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Content Similarity: What? </h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">

<div style="width:100%">
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/topic.png">
<h3 style = "color:white; ">1. Posts to vectors</h3>
</div>

<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/protractor.png">
<h3 style = "color:white; ">2. Calculate Similarity</h3>
</div>

<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/connection.png">
<h3 style = "color:white; ">3. Aggregate to network</h3>
</div>
</div>

<div style="width:100%; ">
<h2 style="text-align:right; margin-top:10em; color:white;">Focused on author content, <br> Not author, or content individually</h2>
</div>

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Content Similarity: Why?</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">

<div style="width:100%">
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/connection.png">
<h4 style = "color:white; ">How authors relate short term</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/singer.png">
<h4 style = "color:white; ">Focus on behavior rathern than topic</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/dominoes.png">
<h4 style = "color:white; ">Examine cascades and information diffusion</h4>
</div>
</div>

<div style="width:100%; ">
<h2 style="text-align:right; margin-top:10em; color:white;">Middleground between <br>author and topic focus
</h2>
</div>

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Content Similarity: How? Author Vectors?</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:2em; height: 5px; width:76%">
<br>
<h3 style="text-transform: none; text-align:left; color:white;float:left; margin-bottom:3em"> Train author-time embeddings?</h3>

--
template: main
background-color:#AF002B

<div style="width:100%; float:left">
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/noise.png">
<h4 style = "color:white; ">Tweets are short</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/stopwatch.png">
<h4 style = "color:white; ">Twitter cycle is fast</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/sad.png">
<h4 style = "color:white; ">Not enough data for meaningful vectors</h4>
</div>
</div>

---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Content Similarity: How?</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">
<br>
<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;float:left">N-gram Spikes</h2>

<div style="width:100%; float:left">
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/spike.png">
<h4 style = "color:white; ">1. Identify n-grams more frequent at t than t-1</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/collaboration.png">
<h4 style = "color:white; ">2. Tie from A to B if B uses n-gram in window after A</h4>
</div>
<div style="width:33%; float:left">
<img style="width:75%"  src="../figures/net_text/news.png">
<h4 style = "color:white; ">Adjust threshold and time window to capture current events</h4>
</div>
</div>

<div style="width:100%; ">
<h2 style="text-align:right; margin-top:10em; color:white;">Capture cascade timing <br>rather than direct shares
</h2>
</div>

---
template: main
background-color: #01BAEF

<div style="width:95%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">Content originates with a few, diffuses, then peters out</h4>
</div>
</div>]

<div style="width:95%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; text-align:center;">Insular information sharing environment</h4>
</div>
<br>
<h2 style = "color:white;">Weak Clustering</h2>
</div>

<div style="width:95%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; text-align:center">Information can originate anywhere in cluster</h4>
</div>
</div>]
---
template: main
background-color: #01BAEF

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Content Similarity: Insights</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<center><div style="width:100%; ">
<div style="width:50%;">
<img style="width:30%; margin-bottom:0px; "  src="../figures/net_text/centrality.png">
<h2 style = "color:white;">Centrality </h2>
</div>
</div></center>
<div style="width:100%; margin-bottom:1em">
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;">High In-Degree: Not early adopter</h4>
</div>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;">Low In-Degree: Either early adopter, or non-adopter</h4>
</div>
</div>
<div style=" float: left; width:100%">

<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">High Betweenness: Perpetuate during lull, promoter</h4>
</div>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">High out-degree low in-degree: Early adopter, initiator</h4>
</div>
</div>

---
template: main
background-color: #6d288e

---
template: main
class: middle
background-color: #6d288e

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Political Journalists</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<h2 style="text-transform: none; text-align:left; color:white;">What is the effect of Twitter friendships on journalist framing?</h2><br>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<h2 style="text-transform: none; text-align:left; color:white;">How do journalists interact on Twitter?</h2>
<h4 style="text-transform: none; text-align:right; color:white;">Do they seek novel information, or confirmation?</h4>
<h4 style="text-transform: none; text-align:right; color:white;">How selective are they for propagation of information?</h4>

---
template: main
background-color: #6d288e

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Political Journalists</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<center><div style="width:100%; ">
<div style="width:50%;">
<img style="width:30%; margin-bottom:0px; "  src="../figures/net_text/centrality.png">
<h2 style = "color:white;">Effect of Twitter on Content</h2>
</div>
</div></center>
<div style="width:100%; margin-bottom:1em">
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;">1. Identify bursty ngrams each day for 3 months around election</h4>
</div>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white;">2. Construct network: Ex] Tie from A to B if B uses "anti-choice" within an hour after A. </h4>
</div>
</div>
<div style=" float: left; width:100%">

<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">3. Compare to follower and retweet networks (Future)</h4>
</div>
<div style="width:48%; float:left; background-color:#AF002B; border: 5px solid #FFFFFF; border-radius: 25px;">
<h4 style = "color:white; ">4. Network model, p(tie_content| tie_follow) (Future)</h4>
</div>
</div>

---
template: main
background-color: #6d288e

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Political Journalists: Followers</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">
<br>
<div style="width:100%;">
<img style="width:80%;max-height:100%" src="../figures/net_text/journo_follow.png">
</div>
---
template: main
background-color: #6d288e

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:white;">Political Journalists: Retweet</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%">
<br>
<div style="width:100%;">
<img style="width:80%;max-height:100%" src="../figures/net_text/journo_rt.png">
</div>
---

template: main
background-color: #6d288e

---
template: main
background-color: #F2DC5D

<h2 style="text-transform: none; text-align:left; margin-bottom:0px; color:black;">Acknowledgements and Thanks</h2>
<img class="plain" src="../figures/line.png" style="float:left; padding:0;margins:0;margin-bottom:0px; height: 5px; width:76%"><br>
<h4 style = "color:black; ">Work in collaboration with Lisa Singh and the GU Data Lab</h4><br>
<h4 style = "color:black; ">Work partially funded by a MIDAS grant</h4><br>
<h4 style = "color:black; ">Icons made by or adapted from Freepik from www.flaticon.com is licensed by CC 3.0</h4>