Best way to do relationship charts?

[Awoo]

So if you've ever gone into one of my confessionals, I tend to do these relationship charts where I color people's names in green if I'm close with them, pink if they're an easy target, yellow if they're a threat, etc, and give a short comment on them.

I think it's a really good and easy to read way of seeing what one person is thinking about his competitors. But I always feel like what falls short in that format is trying to say what I think of the relationships of other people.

Like for example, if I was a spectator, or I was a player trying to map out people's motivations and I wanted to do something similar, to try to put everybody's relationships, alliances, and personal agendas into one easy to read document, is there any known good format for that?

[Awoo]

yeah... the reason I made this thread is because the nodes-edges thing is kinda hard to scale, especially from the spec POV who sees all

[Awoo]

?????????????????????????????????

[Awoo]

Bumping this thread with more stuff because there is still no solution.

https://www.tapatalk.com/groups/survivo ... -t494.html

Credit to papa zito for this contribution. It looks okay but it's hard to view all the different images. Focusing one person at a time is a good idea because it brings the amount of information you need to view at a time from a factor of n^2 down to n, but it's still not quite enough. It also has the (same) problem that we can't view multi-person alliances easily.

[Awoo]

https://www.mafiascum.net/social-games/ ... opic=160.0

This is my attempt at making graphs scale, from when I specc'd survivor equestria. I did my absolute best to make graphs readable by using super thick, transparent, and coloured edges, and I even did a lot of manual adjustment to maximize how readable the graph is. Despite my best efforts, in only 1 episode, two out of three tribes were almost completely unreadable, and the third tribe only made sense because it was so simple. I don't want to try to make one of these in a merge situation.

[Awoo]

A chart is just a way of aggregating the data that players give. If confessional information is sparse in a game, a good quality chart can at least serve to aggregate the information that the specs do have. You can do whatever you want to get data, be it spec questions, plain old reading confessionals like I did in equestria, reading into votes, or even taking your own creative liberties and guessing what the people who don't post in their confessionals are doing through SPV and voting patterns. It doesn't really matter imo.

The problem I'm concerned with isn't how to get the data, it's how to display the data. All the social network visualization software out there is targeted towards finding the big picture or key actors in large, complex, and often rather sparse networks, whereas the spectator of an LSG wants to soak up all the juicy gossip of very small but very dense social networks. As far as I know, there is nothing out there that addresses this problem effectively.

The traditional approach to dealing with small, dense networks (classrooms) is using sociograms, but those don't really go deep enough for an LSG spectator, plus they rely too much on spec questions (who is your closest ally? ...every single round.)

[Awoo]

No, haha, who does that? We're talking about the spectators who are trying to make sense of all your vibing.

Here's an new option, it's a thing called NodeTrix. Here's the paper. If it's tl;dr, scroll all the way down to the bottom page and then zoom in. They give a lovely image as an example. Basically this system combines the two major ideas posted so far in this thread, which is node-edge diagrams and spreadsheets. It finds the communities in a network, and then represents those communities as matricies. I can't seem to get it to work, perhaps it is because their documentation looks like some sort of postmodern meme format.

Spoiler: when you {2}

Here's an example of what someone did with it:

Spoiler: test2

Thinking of a merge situation for example, I can see something that combines a graph with a special tool for detailed community visualization being effective at understanding the big picture of a game [Side A vs. Side B with a couple swing votes] and then getting a more detailed picture of each faction and the interactions between them and amongst themselves using the community visualization tool. [Swing votes are closest with these people, who is the leader of side A/B, etc.]

The downside, as with most node-edge based tools, is that it doesn't seem to handle different types of relationships [friend, enemy, nonmutual, etc] very well.

[Awoo]

This is probably the best chart I've seen, coming from meme playing as Rainbow Dash in Equestria.

https://media.discordapp.net/attachment ... height=619

There's so much that is good about this chart.
1. It focuses on an individual player, which scales the information density from n^2 to n.
2. Not only does it show all of RD's relationships, it shows every group she is involved with, as well as all their members.
3. Displaying enemies is so easy, just don't include them in any circle.
4. Non-mutual / shaky relationships are accounted for as well, look at the dotted line circle with Soarin in it.

Downsides: It does not look easy to make one for every single person in the game!!!

[Awoo]

That's the problem - different players view their connections using different metrics and what you're looking for in part is a method of standardising those connections. An optimal solution would probably have some sort of spectrum for how close players/alliances are to each other.

I think there are three (easily quantifiable) dimensions to relationships in LSG's.

(a) Do I like this person? (Do they like me?)
1. Yes
0. Neutral
-1 No

(b) Do I think this person is threatening, or weak? (Do they think I am threatening or weak?)
1. This person is threatening
0. Neither
-1 This person is weak

(c) Based on the two previous questions (or not), do I plan on keeping this person in the game? (do they plan on keeping me in the game?)

1 I would fight against a vote on this person
0 I would not fight against a vote on this person
-1 I would be enthusiastic about a vote on this person

These dimensions go both ways, so they can all be either mutual or non-mutual.

(X) --- celi I love you ---> (Y)
(X) <--- regi u suck ----- (Y)

would be
(X) ==[1] => (Y)
(X) <=[-1]== (Y)

On (a), I don't even want to get into trying to chart how much someone "trusts" someone else. In my experience, that is incredibly subjective and changes on a dime. I feel like you have far better odds just recording how much people say they like each other, since players tend to say that they like people that they trust and vice versa. So in this sense, (a) is a bit of a fuzzy mix of {social compatibility, game compatibility, trust} and all the other hidden factors in the human heart that make someone worth liking or not. People also tend to post who they like and dislike in their confessionals, so you can get away without relying on spec questions for this one (unless the cast really sucks)

(b) is a game related question, and helps specs understand the "perception is reality" that's happening in the game. Players tend to answer this unprompted, and in a way that is consistent with the way that they end up acting.

(c) So here's the problem from a spec POV, it's easy to get the answer to question (a), because people post that all the time in their confessionals, and if they don't you could post targeted questions. It's easy to get the answer to (b) for the same reasons as (a). But it's not possible to get the answer to (c), because LSG's are complex and sometimes the players themselves don't really know what they want. Sometimes they will say they want to do something the whole game, and intend to do something the whole game, and then never actually do it (read: every single one of my confessionals). Plus, people's plans change based on complex forces that don't nicely fit into data points.

Ultimately, the information that we want to chart is (c), but it's not possible to get a perfect representation of that. So our options are we try to guess at the answer to (c), or we can just stick to charting what we know in (a) and/or (b). Personally I am inclined towards the latter, since some of the fun of being a spec is looking at the gamestate and guessing what will happen next. Or at the very least, having a decent understanding of the gamestate so you can formulate your own hot takes.

To get back to Klick, I believe that ideally you would have some sort of spectrum where you could read everybody's mind better than they can read it themselves and determine exactly how close they are to each individual person in the game on a continuous scale of -1 to 1 or what have you, but I don't think it's practically feasible. At the moment I think the most practical thing to do is get the answers to yes/maybe/no questions and keep track of them that way.

[Awoo]

It's not that complicated, really.

Here's a person that I hate, but I want to drag to the end because they're a goat:

(a) (HERO) --- dislikes ---> (VILLAIN)
(b) (HERO) --- sees as weak ---> (VILLAIN)

Here's a person that I like, but need to cut because they're a social threat

(a) (HERO) --- likes ---> (ALLY)
(b) (HERO) --- sees as threat ---> (ALLY)

Given this raw data and a little bit of confessional reading, we can come to the (subjective, potentially inaccurate) conclusion that:

(c) (HERO) --- would fight against a vote on ---> (VILLAIN)
(c) (HERO) --- would not fight against a vote on ---> (ALLY)

And this is where those personal nuances come into play. Depending on hero's complex nuances, he may or may not want to go to the end with villain.

In general, like/dislike (a) would have more impact on actual intentions (c) early on, and perceived threat level (b) would have more impact in the late game. But since we're humans and not robots, we can't say that for certain. That's why I think the best thing to do is just chart (a) and (b), then leave the viewer to come to their own conclusion about (c).

[Awoo]

Like sometimes you can trust somebody and work with them but absolutely hate them, and you can really like people that you need to throw under the nearest bus.

Wait that happens?

People hating each other but going to the end together: Regi and Deoxys

People liking each other but needing to cut the other for being a threat to win: literally every game in history

[Awoo]

I think there's a little bit of a problem with that scale there, Awoo. Some people are a lot more nuanced than that. Like sometimes you can trust somebody and work with them but absolutely hate them, and you can really like people that you need to throw under the nearest bus. How do you quantify those situations? How do those display differently than the traditional like=trust dynamics? Or do those show up the same?

However here's what she meant:

Hero, Villain, and some other people are being threatened by a big alliance, lead by Cult Leader and need to band together to stop them. Hero has no intentions of working with villain in the long run, and really they don't like each other all that much, but they need to stay together in the short term to survive. So seeing only the following data:

(HERO) -- dislikes --> (VILLAIN)
(HERO) -- isn't thinking about the threat level of --> (VILLAIN)

would not lead you to the conclusion that

(HERO) -- would oppose a vote on --> (VILLAIN)

Which means it is also necessary to somehow communicate players that are grouped together for nontrivial reasons, such as voting blocks, majority alliances on a tribe, outcast alliances, etc. From there you could view the dynamics inside the voting block and reasonably predict that hero is only working with villain for the short term.

[Awoo]

This is a seemingly dead project, and it's not written in english, but the pictures look cute. Posting just for inspiration.

https://github.com/xinyzhang9/friend-circles

[Awoo]

some good reading

https://www.cs.cornell.edu/home/kleinbe ... k-ch05.pdf

[Awoo]

Interesting tech to keep an eye on

https://github.com/penrose/penrose