Print

Print


*****  To join INSNA, visit http://www.insna.org  *****

Hi Maurice and Derek,

We've had some success using supervised trainers to detect speech acts
and argument strategies in tweets. We're getting pretty reliable
results from the trainers (>.75 accuracy) but by no means perfect.
We'll talk about speech act detection at CSCW (paper's available here:
http://share.iit.edu/handle/10560/2884), and I can share a working
paper about argument strategies off-list if you're interested.

For each, we did have to use 3-4 rounds of human coding to get labeled
data that could reliably train the automated classifiers. We've used
existing tools (MALLET and WEKA) to run the training and classifying.
My frustration with much of the SNA + NLP intersection is the overhead
and getting started, so we stuck with Twitter (easy data to get) and
existing machine learning tools to get started.

Libby
-- 
Libby Hemphill
Assistant Professor of Communication and Information Studies
Illinois Institute of Technology
http://www.libbyh.com

On Mon, Nov 5, 2012 at 5:55 PM, Maurice Vergeer <[log in to unmask]> wrote:
> ***** To join INSNA, visit http://www.insna.org ***** Dear Derek
>
> I welcome your post on the SOCNET list on NLP of social media texts, because
> I want to dive into the world of NLP.
> I see the great value of NLP. At the same time I am wondering what the
> quality is of NLP on everyday language on sociale media. Particularly
> regarding Twitter data I have some doubts as to the quality of the
> NLProcessing. Having coded tweets using trained students, it showed that
> particularly when a tweet was to be coded isolated from the discussion
> thread, establishing the meaning of the tweet was quite difficult and thus
> unreliable. Furthermore, Twitter is full of spelling and grammar mistakes,
> abbreviations, shortened URLs, etc etc.
> I can imagine the case is different for Facebook, having more space to
> convey ones thoughts than on Twitter. However collecting Facebook data is
> more difficult than Twitter data.
> I guess you already ran  some initial tests? If so, would you be willing to
> share some initial findings on the quality?
>
> best regards
> Maurice Vergeer
> Maurice
>
>
>
> On Mon, Nov 5, 2012 at 11:31 PM, Derek Hansen <[log in to unmask]> wrote:
>>
>> ***** To join INSNA, visit http://www.insna.org ***** Hi all,
>>
>> I'm giving a guest lecture in a couple days on the intersection of Social
>> Network Analysis and Natural Language Processing, particularly as applied
>> toward the analysis of social media datasets. I know there's been a lot of
>> activity lately at the intersection of SNA and NLP and I'd love to see a few
>> pointers to some good quality work in this area.
>>
>> Additionally, I'm hoping to flesh out a few high-level approaches for
>> merging the two in meaningful ways. Here are a few examples:
>>
>> 1) Use content-based factors (extracted via NLP techniques) as well as
>> SNA-based metrics as independent variables that predict an outcome of
>> interest (e.g., quality of content, such as was done in Agichtein, Castillo,
>> et al. 2008).
>>
>> 2) Use NLP to help define the edges in a network (e.g., "link polarity" as
>> performed by Kale 2007).
>>
>> 3) Use a 2-step filtering process:
>> 3a) Use SNA to identify network clusters and then use NLP on the corpus
>> created by those within each cluster (e.g., Marc Smith's graphs on the
>> NodeXL Graph Gallery where keywords are overlaid on the network clusters)
>> 3b) Use NLP to identify subsets of "relevant" content whose authors are
>> then analyzed via SNA.
>>
>> 4) SNA helps in disambiguating words (e.g., when one network cluster uses
>> the term "jaguar" they typically mean the sports team, while another network
>> cluster typically means the car).
>>
>> Other thoughts on high-level strategies would be welcome as well.
>>
>> Regards,
>> Derek Hansen
>> Brigham Young University
>> _____________________________________________________________________
>> 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.
>
>
>
>
> --
> ___________________________________________________________________
> Maurice Vergeer
> To contact me, see http://mauricevergeer.nl/node/5
> To see my publications, see http://mauricevergeer.nl/node/1
> ___________________________________________________________________
> _____________________________________________________________________ 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.



-- 
Libby Hemphill
Assistant Professor of Communication and Information Studies
Illinois Institute of Technology
http://www.libbyh.com

_____________________________________________________________________
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.