1 of 35

DH Capstone

Social Media Analytics

Marissa Clifford, Stephanie Wong, Swati Katta,

Jonathan Solichin, Joanna Chen,

Thanks to Professor Todd Presner, Professor Francis Steen, Yoh Kawano, and David Shepard

The DH Capstone Social Media Anaytics research documentation. Research began with Stephanie Wong, Marissa Clifford, and Jonathan Solichin in Winter 2014. More information on the background of the research can be found in the Google doc, “CSA Social Media Module” https://docs.google.com/document/d/1egONgsV_zwwC44ohDO9tOAmNvwSFDBFrzKqUqkWZkGs/edit

This powerpoint was presented in class as a combination of our presentation and documentation for Spring 2014. The group consisted of Marissa Clifford (graduating senior, Art History, DH minor), Stephanie Wong (graduating senior, Communications, DH minor), Swati Katta (graduate student, Computer Science, GSR), and Joanna Chen (graduate student, Library and Information Science, DH Certificate). Special thanks to Professor Todd Presner (listed instructor), Professor Francis Steen, Yoh Kawano, and David Shepard. The final presentation that was given for DH was created on Prezi and can be accessed here: http://prezi.com/kluuirlfzfol/dh-capstone-social-media-analytics/

-Joanna Chen

2 of 35

The Big Question Then

Develop World Watcher: A Tool for Automatically Discovering and Recording Events as they are Represented in Social Media

Index Users
Compare elite and social media

3 of 35

Adapting Our Methods and Goals

Find reliable and sustainable method for identifying and monitoring Index Users
Gathering data and trying new approaches
Trend toward visualization

4 of 35

The Big Question Now

Agenda Setting
Primacy in the News
Analogical Reasoning
Explicit Causal Reasoning
Duration and Persistence
Sentiment Analysis

5 of 35

Background: Boston Bombings Research

1) Evidence - What Happened?

2) Explanation - How did it happen?

3) Event Surgery - Why did it happen?

4) Accountability - Who is to blame?

5) Planning - How to prevent it?

6 of 35

Towards a Sentimental Dictionary: Emotional Intensity

0

1

2

3

Highly Subjective Editorial

Completely Neutral Reportage

Just The Facts

• All CAPS

• Expletives

• !!-!!!

7 of 35

Data

Twitter
Reddit

8 of 35

Twitter Data

Twitter’s API was very useful in obtaining the data from the live stream
Initially, we decided to find some index users who we could watch for tweets related to important events
These index users were selected on the basis of several parameters

9 of 35

Index Users

Sample the live stream for tweets and get the twitter handles of their authors
Determine if the author has the following:

– High Klout score

Score obtained from Klout API

– High number of followers

– Geographic location : United States

10 of 35

Twitter Timelines

We obtain the timelines of each of those index users and analyze whether they are talking about important and interesting events.

11 of 35

Main issue : SPAM!!!

12 of 35

Reddit Data

Another source of news and information about events
More anonymous unlike twitter which is focused on the user
Could help in filtering tweets from twitter
More organized

13 of 35

Using Reddit to filter Twitter

Fetch Reddit headlines

–Reddit API is not useful

Use selected keywords to filter live stream of twitter

14 of 35

Learning Curve

used D3 examples
replaced with our dataset
altered the json and html codes
created our own tutorial

15 of 35

Visualizing the Data

16 of 35

Picking a visualization

17 of 35

Using Anvil

Create a local server (best). For example (mac): download http://anvilformac.com/. an icon should pop up in your menu bar that looks like an anvil. click it. press the plus button. Add the folder in which you have saved all your files.

MODIFIED CODE: http://pastebin.com/ZZbEbfhV

Replace the content of index.html with this code. To clarify, you must simply copy the section in the modified code above from the part that says “*begin modification*” to the part that says “*end modification*”. Then go back to your original index.html file in your text editor. Find the line that says “d3.json("flare.json", function(error, root) {“ and replace that line with the portion you just copied. Save the file.

Download this file and place it in the same folder: http://sandbox.idre.ucla.edu/up206b/2014/dh199/jssolichin/ArchiveRedditHeadlines.csv

Refresh browser (or click the .dev file in anvil to see it open in your browser) and see Reddit visualization! -Stephanie Wong

18 of 35

ONTOLOGIES

Ontologies:

After spending a quarter collecting data from Twitter and Reddit, learning visualization tools such as D3, and discussing how to analyze and visualize news on social media, we decided to focus on the recent UC Santa Barbara shooting. We were interested in what conversations social media users were engaging in regarding the shooting, whether and how that conversation changed over time regarding the topics put forward, and whether there was a difference compared to elite media. To create the sunburst visualization, I created a list of terms based on what I found to be the most popular topics in reading through news articles. I then created a hierarchal structure in which terms either fell under other terms as synonyms or as sub-topics of the original topic. Of course, as with creating any thesaurus, the topics, terms, and structure could benefit from more in-depth research and study of the content and how to best break down the topics. -Joanna Chen

19 of 35

20 of 35

Here, we used word clouds to visualize the types of language used and trending topics discussed between elite and social media. The top cloud was created using closed captioning from broadcast media archived on Newsscape, and the bottom cloud was derived from Twitter data -- all focused on the topic of the UCSB Shootings. It is clear that more analytical and socio-cultural themes were present in the social media -- hashtags like YesAllWomen, and discussions of Judd Apatow’s new film Neighbors and how media perpetuates normative roles of feminity and masculinity. When we juxtapose this to the top cloud, we see much more discussion of the factual unfolding of the event, particularly with interactions between citizens (suspect, victims), and those responding to the scene (police, reporter, deputies...etc.).

21 of 35

Visualizing the Ontology

Parse all tweets and get word frequency count
Put frequency count into the ontology hierarchy
Visualize

(Each outer radius represents subdivision of the inner radius, starting with all tweets)

22 of 35

Visualizing the Ontology

eg.

We can see that misogyny makes a lot on the topic of women, #yesallwomen makes up a lot on the topic of misogyny, and violences makes a lot on the topic of #yesallwomen.

23 of 35

Visualizing the Ontology

eg.

We can see that killing makes a lot on the topic of shooting, rampage makes up a lot on the topic of misogyny.

and so forth...

24 of 35

Demo

http://sandbox.idre.ucla.edu/up206b/2014/dh199/jssolichin/sunburst3/

25 of 35

Issue with this viz

Tweets are combined into one document, and thus the graph does not represent # of tweets, but the number of mentions of�the topic.

26 of 35