Extracting Communication Networks - Fundamentals

[HyunkuKwon]

Post questions here for one or more of our fundamentals readings:

Fortunato, Santo. 2010. “Community Detection in Graphs.” Physics reports 486(3-5): 75-174.

Borgatti, Stephen P. 2005. “Centrality and Network Flow.” Social networks 27(1): 55-71.

[egemenpamukcu]

I have two general questions:

1. In the context of content analysis, what relationships can we capture with community detection that we cannot understand with topic modeling, clustering or classification? Why would we choose community detection over these methods?

2. What factors should we think of when we decide on what constitutes an edge in our network (or even a node)? It seems like it is more of an art than science.

[william-wei-zhu]

Network analysis seems to be very useful in describing certain patterns or draw associations. I wonder can we detect causal relationship using network analysis techniques?

[RobertoBarrosoLuque]

Question: network theory seems to be broadly applicable to just about any field. A particularly interesting application is the use of graph and network theory to vaccine distribution https://www.wired.com/story/covid-19-vaccine-super-spreaders/  My question is about identifying worth while applications, which framework or set of questions should we ask as content analysts when initiating network analysis?

[chiayunc]

The author has provided different ways of measuring centrality. I wonder if those methods are applicable to semantic networks? or are they more suited for a network where there exist flows between nodes. And what might a semantic network with flows look like?

[xxicheng]

Are there more examples of applying community detection to social mobility research? I am thinking about the possibility of its application beyond social media.

[romanticmonkey]

I wonder if there are studies on how ideas spread from culture to culture, or country to country. If there are, could you please share them with us?

[k-partha]

The reading on centrality and network flow was insightful in its argument that the nature of traffic flow is critical in understanding which centrality measures are appropriate. In practice, how would one classify the kind of traffic flow we are dealing with in our data into the different categories outlined in the reading? (eg. walks/trails/paths)

[lilygrier]

The reading on community detection in graphs presented a very thorough introduction to how graphs work and the different algorithms that can be used to create them. There are benefits and tradeoffs associated with choosing to be more or less stringent in how one defines a neighborhood/module within a graph. Some of the applications involve direction communication (i.e., protein interactions, a Facebook friend network) and some are more indirect (e.g., a network of citations in academic papers). Is there a way to apply network analysis to non-reciprocal/passive information dissemination, such as people "lurking" on Twitter or reading news sources without commenting on them (provided anonymized data with pageviews were available)?

[Raychanan]

My question is about Borgatti's work. Obviously, the various measurements in this paper are not "very" different in terms of performance, although they are designed in very different ways. So, I am a bit skeptical of the measurements presented by the authors. Could you please explain why their performance is so similar?

[ming-cui]

As Borgatti (2005) indicated, a common criticism of social network research is that insufficient attention is paid to network dynamics. I am also facing this issue. In a study, I surveyed group members' perceptions (e.g., power perception) of each other over five time points. I may need a longitudinal social network analysis technique or something to analyze the data, but I haven't figured out an appropriate solution. Is there any available analysis fitting this study? Thanks!

[jacyanthis]

Centrality and community seem to be precisely defined by the human researchers. Can we instead show a neural network a large number of example networks with various properties defined (e.g. vertices A, B, and X are a community) then let it build its own measures of centrality and community? What would we get?

Or perhaps the measures have gotten good enough since 2005/2010 that we don’t need to worry about this.

[Rui-echo-Pan]

Hopefully, you could provide some clarifications on the differences among community detection, topic modelling and the classification related.

[MOTOKU666]

Following @romanticmonkey 's idea, I'm also interested in the specific pathway of ideas spreading across different cultures. Also, the language may naturally affect the circulation and accumulation of idea based on their features.

[zshibing1]

There seems to be a lack of convention on the use of terminology. What are the meaningful differences between community and network?

[Bin-ary-Li]

It feels like cluster learning is a big open question that requires deep mathematical finesse and quantitative knowledge. What does that bode for future sociologists/sociology students who want to study network? Aren't they better off started with applied math/statistics instead of sociology?

[theoevans1]

What kinds of relationships or interactions can be conceptualized as networks? Much as prediction models can be used for questions that aren’t themselves predictive, are there surprising types of questions that network analysis methods can be used for?

[jinfei1125]

I find that social network analysis is borrowing many ideas from topology. So I am wondering that in content analysis, what are the most prevailing topology method used in our textual data analysis?

[hesongrun]

In the paper Borgatti, Stephen P. 2005, the authors introduced four kinds of centrality measurement: closeness centrality, betweenness, eigenvector centrality & degree centrality as well as their strengths and weaknesses. I find it fascinating to link these mathematical concepts with underlying appropriate social game processes. This is the power of computational social science!

In the real world, we are often given large amount of unlabeled data. I am wondering if there are some efficient ways to construct networks with unsupervised methods that can represent the underlying community or clusters? Thanks!

[dtanoglidis]

Network analysis is an exciting field and I had fun skimming through these review papers. A question I have that is not explicitly mentioned is the following:

There is no much discussion about uncertainties: although in some networks (e.g. phone communications) the edges and nodes are easily to be measured, in other cases, especially of social networks (strong vs weak ties) these can be harder to define and may depend on the assumptions of the research(er). Is there a way to quantify these uncertainties and propagate them to e.g. the centrality measures of the network and thus give some "error bars" in these quoted values?

[jcvotava]

I am still not sure about using networks to describe communication patterns vs. networks as used to describe semantic relations and linkages. What are the differences between these kind of network constructions?

[mingtao-gao]

The paper mentioned about signed and unsigned networks. There are already related package to build these networks in R, however, in practice, how do we decide if we should choose signed or unsigned networks for our own project? Should we just try both and see which one is more suitable?

[toecn]

What are interesting ways of analyzing the evolution of networks over time?