***** To join INSNA, visit http://www.insna.org *****
I'm not sure if what I'm going to say answers your question, but I was
wondering whether you could consider focusing on all the tweets of
your dataset and measuring "happiness" on the text of each tweet,
independently if it is a reply, a mention, an RT or anything. Of
course, in this way, you don't have a proper network, unless you want
to filter out just the RTs and the mentions, in which case you will do
have a directed graph, or you would be pleased to consider some sort
of a two-mode network (for instance, Twitter users and hashtags or any
sort of Twitter "events," which may be automatically extracted from
your Twitter data by searching for specific terms or keywords or doing
Topic Modeling to detect their important categories).
I'm doing some sort of similar analysis on a big dataset of tweets
about the issue of Global Refugees during the last 8 months (since
October 2015) and although it is an ongoing work I might send you the
link to the Beaker Notebook where you could see our preliminary
computations and visualizations. As a matter of fact, in what concerns
hedonometrics, we're simply doing sentiment analysis (through NLTK's
TextBlob) and we are measuring the scores of polarity and
On Sun, Jun 19, 2016 at 8:51 PM, Ian Cero <[log in to unmask]> wrote:
> ***** To join INSNA, visit http://www.insna.org *****
> Hello SOCNET,
> I hope this message finds you all well and thank you in advance for your help. Briefly, I am
> new to social network analysis, so I am seeking some guidance on edge criteria and other
> best practices.
> For a project on mood assortativity on Twitter, I have been using Bliss et al.'s (2012) paper
> on happiness assortativity in that network as a template. There, the authors define an edge
> as a reciprocal-reply and proceed to analyze the giant component produced for a week's
> work of tweets.
> What makes me pause is that the ratio of replies to observed tweets currently coming
> through Twitter's streaming API is about half as large as what was reported in that paper,
> whose tweets are from 2008. Likewise, while those other authors report about 10,000
> vertices in their giant component each week, I am finding about 60 in mine (perhaps a
> reflection of a much sparser reciprocal-reply network).
> So, my questions are as follows:
> 1) Could a change in Twitter usage account for a discrepancy of that size between my
> results and the previous paper? For example, are people simply using the reply function less
> and @mentions more?
> 2) Are their any publications on the best edge criteria to use on Twitter for a case like mine?
> 3) I think my giant component is too small to match the kind of analysis I hope to do
> (ERGM). Is it okay to analyze the entire network I have, even if the components of that
> network are disconnected?
> Again, thanks for any thoughts you have,
> Bliss, C. A., Kloumann, I. M., Harris, K. D., Danforth, C. M., & Dodds, P. S. (2012). Twitter
> reciprocal reply networks exhibit assortativity with respect to happiness. Journal of
> Computational Science, 3(5), 388–397. http://doi.org/10.1016/j.jocs.2012.05.001
> SOCNET is a service of INSNA, the professional association for social
> network researchers (http://www.insna.org). To unsubscribe, send
> an email message to [log in to unmask] containing the line
> UNSUBSCRIBE SOCNET in the body of the message.
SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to [log in to unmask] containing the line
UNSUBSCRIBE SOCNET in the body of the message.