For my Gephi practicum, I took data from the Women Writers Project Cultures of Reception project (forthcoming) to visualize the relation between 18th/19th century women writers and the literary journals that reviewed them. I looked at five different journals and made a list of all of the women they reviewed. These comprised the nodes in my graph. The edges are directed (with the journal as the source and the author as the target), and I recorded multiple instances of one journal reviewing one author, which ultimately impacted the edge weight. If you were to look at my [edges].csv file, you would not that I also differentiated by year (although that doesn't show up in my final visualization). You can see the resulting graph here:
Creating the csv files and loading it all up into Gephi were both relatively easy, but I found that the subsequent part was much more difficult to conduct without extensively theorizing the materials I was working with. The WWP has extensive information on all of the writers and journals in the database (nodes), and is currently collecting information on the nature of the reviews (edges). However, I wasn't able to use all of this information for my visualization, since I only had limited time to complete the assignment. I can imagine using the information on the journals (such as whether they had conservative or liberal political leanings, where they were published, etc.) as attributes and organizing the graph accordingly. This may have helped me to understand something about why Gephi found certain communities or why certain authors were reviewed so frequently.
This is a perfect example of why a researcher's desire to use digital tools should always come from an interest in the research object, rather than a desire to use the tool. I'm sure if I spent more time working with the Cultures of Reception material, I would have found something that would benefit from a network visualization (literary reviews seem like good candidates in a knee-jerk sort of way, which is why I picked this topic in the first place). However, without a preexisting notion of what the network might look like, it's very hard to actually create a visualization that makes any substantive argument about the data.