In game chat dump from 631 711 replays and 23 929 players
-
here is a folder of 23 929 user ids and what they have said in the last 631 711 FAF games.
this was parsed with askaholics tool and some custom python sciripting
https://forum.faforever.com/topic/1551/faf-scfa-replay-parser-libraryFrom this data you can infer some useful information.
UserID is used, because unlike username, it does not change. The "notify" messages have been excluded.
like which players get fed the most mass
who uses the most AI taunts
who is the most toxic (probaly best to define toxic to a script, hard to do)https://drive.google.com/file/d/1hKd671xIQoE92zB27PCdjB4uqBucu39g/view
Unzipped size is ~ 120mb
all the user ids are in one folder, windows may struggle.
I plan to do more replay analysis including node graphs of who plays with who, who plays on what map,
correlation between:
map and rating
country and rating
map and countryI also plan an analyse the notify dataset (not included in this chat dataset) and get information on upgrade timings, of acu also per faction and all of them per rating group and map.
any others you can think of.
Please download yourselves and anaylse
-
There must be something wrong with the data because in my case it only includes like 10 lines and I have definitely said more than that.
Or maybe 631711 replays is less than I thought. How far does it go back?
-
@femtozetta there is nearly 15 million total replays
-
So this is about 1 year?
-
I would kindly request you not to analyse peoples chat messages and then publish names out of that.
People did not consent to this and FAF could get in trouble for such a thing.
For things like who played against whom, replays are a bad start. Its easier to scrape the api and put it in a graph database (SQL is incredibly inefficient on this particular kind of queies)
-
@brutus5000 Arent all the replays and their chat histories searchable by username? Like, publicly?
-
Your camera doesn't stop you to film other people. Yet if you upload a video to YouTube that is not allowed.
Similar to IRC. You can keep a chat log of everything. But publishing it is a grey area.Just to get this right:
- You are allowed to analyze this data for private purposes.
- You are allowed to share insights (e.g. see the funny cloud diagram of most used words in chat)
- You are not allowed to analyze single person behaviors or do a social rating and publish this (or basically do anything that relates back to a single user)
If I see it on the official FAF platform anywhere I will remove it and give that person a warning (or have moderation do it).
However, I cannot prevent you from publishing it elsewhere and if you do so and it falls back to us legally in worst case we'll have to take other measures whatever they might be (e.g. take down the replay vault as the final action if no other solution).That's why I'm asking kindly for now. Just let's stay out of trouble and everybody can still have fun.
-
2 questions from an amateur:
- Could you upload a bigger dataset as well?
- Is there some webservice or database or something where I can translate userid to username easily?
-
The chat logs are already published in the vault so I don't see how this is any different tbh
-
You don't see a difference between browsing through chat logs manually and mass-profiling single users and publish the results?
-
I mean, it's just presenting public information in a more organized and readable way. It's the same difference as instead of looking at rating changes from each game in the replay vault there is a tool that shows you rating changes of each user every day and other similar tools eg. Kazbeks tool that allows you to see what map a specific user plays. Yes, in theory, it can be used in a harmful way, idk shaming someone for thrash-talking people or swearing at them or something.
-
Not to mention commands like top words that achieve the same thing but on a smaller scale.
-
How would FAF not be liable for some information issue here?
They :
- have made replay information public for everyone
- have made a parser to allow you to, uh, parse this information and even included instructions on how to use it
No idea about Europe but in the US there is a liability doctrine that doesn't let you just give a person tools, say "don't do that bad thing with the tools" and then wash your hands when you put zero effort into making it difficult to actually do said bad thing.
The only thing FAF hasn't done is give you step-by-step instructions on how to download replays from the vault to then use the tool.
I mean I don't get the issue in the first place, do people have legal ownership over the words they write in game or something? Wouldn't this already make the replay vault a "legal liability" unless you requested consent before publishing any replay?
Also, "You are not allowed to analyze single person behaviors or do a social rating and publish this (or basically do anything that relates back to a single user)" isn't this essentially what moderation does? Don't report results get reported back to the person that made the report? That's a publication of the analysis of a singular person's behavior.
-
I dont mean to have caused any legal trouble here, I was just interested in some data analysis. I should probably have started with things other than text chat first and gauge a response but text chat was the easiest for me to parse and make sense of.
from my perspective the information is already highly available in the replay vault publicly.
I do understand open source data can become sensitive when massed together.
Perhaps we need a disclaimer that replays and all information contained are available publicly? along with name history and rating history and anything else, ect? It has always been very obvious to me that they are but to others it may not be?
-
Legality aside there are clear morality concerns. A LOT of miscellaneous personal information is public on the internet if you try hard enough to search for it, but collecting and publishing it on public forums is not really ok. Argument that "it's already public" only stands up if you delve into technicalities. And if we had some certain moderator still active giving them this kind of idea would likely result into a mass mega ban or a big drama fest.
-
Everything is open assuming good will:
a) don't misuse the data
b) don't cause performance issues on the serverAs long as everybody behaves we're good. If I see misuse I'll shut it down / make it unavailable to the public.
So far no lines where crossed, but I hope I made my point clear where the red lines are.So @Nooby you did not cause any trouble yet. I just tried to proactively step in before things go in the wrong direction.
-
APOLOGIES THIS POST IS IN CODE FORMAT - it was the only way I could show my post while keeping the tabination of the word & count tables..
Currently learning python and decided to play with and analyse this dataset just out of curiosity. It contains 20,405,216 words, spread across 23,929 files (representing that number of games) for a total data size of 112+MB I removed the 2,760,185 non English words as I am only able to speak one language. So that's all the Russian, German etc words removed. So there are 17,645,031 English words remaining, let's look at these. None of the following proves anything, I just thought it would be interesting to have a look. What are the actual most commonly used words? WORD COUNT ---------- ------- to 1464477 sent 1305978 mass 743675 energy 667411 you 274816 me 263903 i 245113 give 223129 gg 179505 can 151937 the 131350 Nothing too suprising there. Let's look at some other word counts now. Other words of note very commonly used: WORD COUNT ---------- ------- air 92538 units 83868 lol 66198 unit 60286 t3 51683 need 49956 dont 48234 why 30133 help 25714 How friendly are the games? WORD COUNT ---------- ------- pls 63084 gl 41712 hf 39946 plz 26900 ty 26191 nice 26087 please 26005 glhf 19296 thx 16672 sorry 15812 thanks 8638 sry 5498 How toxic are the games? Actually, not as much as I might have worried.. WORD COUNT ---------- ------- fuck 21997 shit 19482 fucking 16963 frustrating 16911 fucked 6089 ffs 5988 damn 5585 idiot 5494 ass 3068 asshole 800 What about issues in the game? WORD COUNT ---------- ------- lag 13433 re 19892 kick 10327 afk 5564 lagging 5035 eject 4792 lags 4251 How often are the game enders mentioned? WORD COUNT ---------- ------- nuke 28959 mavor 13966 para 6724 paragon 3185 scathis 4679 yolo 4106 novax 1669 yolona 1296 salvation 1047 And the experimental units? WORD COUNT ---------- ------- spider 6160 monkey 4272 gc 4031 chicken 2145 fatboy 2340 czar 1874 mega 1650 fatty 1335 tempest 1188 ahwassa 1083 monkeylord 501 megalith 473 ythotha 449 ripper 384 atlantis 379 asswasher 348 colossus 312 soulripper 55 Which races get talked about most? Presumably due to asking for engineers to make Hives and Kennels: WORD COUNT ---------- ------- cybran 13614 uef 13065 aeon 8793 sera 5768 seraphim 1088 (much faster to just type sera!) How are the commanders referred to? WORD COUNT ---------- ------- com 11785 acu 11555 Does playing FAF give you headaches? Because ibuprofen is mentioned 187 times. -
@scout_more_often Dont apologize! This looks amazing dude! Could even make a graphic out of this data! a FAF interesting Chat Facts sheet!
-
Would be an interesting fun-fact news. The ibuprofen thing is funny.
-
How does one find their user-id?
