In a nutshell, the authors hope to get some insight into whether a myth is based on fact by seeing whether the social network of characters in the myth looks more like a real social network or like the social network in a work of deliberate fiction. For instance, the social networks of the Iliad and Beowulf look more like actual social networks than does the social network of Harry Potter. Real social networks follow a power law distribution more closely than do social networks in works of fiction.
I vaguely remember reading about some astronomer's using Homer's descriptions of constellations to pin down dates for various events in the Odyssey. Perhaps it was this PNAS paper by Baikouzis & Magnasco?
It seems however that an accurate historical account might have a suspicious social network, not because the events in it were made up but because they were filtered according to what the historian thought was important.
Indeed. Although I suspect this form of Narrative Bias would be less of a problem with Beowulf, the Illiad, etc., because the composers of those tales, and their audiences, had less exposure to the way fiction is "supposed" to be.
I would like to see someone do similar analysis for contemporary non-fiction. I prefer authors who tell a sprawling, tangled, less narratively driven story (Keay, Mann, Lewis, and Mukherjee come to mind) to one that fits a more conventional structure. It's difficult for me to accept a story that takes place in a high causal density environment and yet somehow only a couple of people have any agency.
Nate Silver apparently feels the same way:
When we construct these stories, we can lose the ability to think about the evidence critically. Elections typically present compelling narratives. Whatever you thought about the politics of Barack Obama or Sarah Palin or John McCain or Hillary Clinton in 2008, they had persuassive life stories: reported books on the campaign, like Game Change, read like tightly bestselling novels.
— The Signal and the Noise, p. 59
Those are not the books I'm interested in.
NLP is very much not my area, but I can't help but wonder about automatically generating some sort of metric about this for a book. Count up how many proper names appear, and build a graph of them based on how closely together they are linked in the text. (That is, two names appearing often in the same sentence are more closely linked than two which appear in consecutive paragraph.) Perhaps simply looking at the histogram of name occurrence frequency might give you some preliminary ability to separate books into "realistic social structures" and "fiction-like social structures."