Potential for AI interpreted trade logging

#1

(Wasn’t sure where to put this but seems as good a spot as any)
I’m looking into the possibility of training a neural network to interpret live nexus chat for trade offers, and converting the messages into trade proposal data –
Meaning, what item(s) they are buying/selling and what they want to trade for it (if listed).
Anyway, I’m reaching out to RealmEye because:

  1. Wouldn’t want to collect chat messages in any “unofficial” way
  2. RealmEye already has some infrastructure set up for listing trade info
  3. RealmEye is already in a position to collect such data

Anyway, I wouldn’t want to anger DECA or anything, and I have no details on the terms of the relationship between RealmEye and DECA, but I just wanted to get a feel if there were any interest and/or info that would be helpful in determining if I should waste my time delving into the AI training and coding or not. Feel free to DM me if necessary. Thanks!

(also, more than willing to explain in more detail what I’m going for)

#2

What kind of model would you even use for the neural network? I feel like using a NN for this is a kind of round-about way to do things.

#3

That sounds utterly pointless. I mean there are two main ways trade. There’s posting on RealmEye’s trade section. And there’s standing around announcing your trade. They are different in a number of significant ways.

RealmEye is better for less common trades such as particular items, as the trade can stay up for days and weeks until someone notices. It’s also better for precise trades because of the limited scope for asking for “offers” or “any pots except xxx”.

Announced trades though tend to be for very common items. Pots most often, but also common gear, consumables. Trades are much more variable, as people in a hurry often just want something for whatever’s clogging up their inventory, often accepting under/overpayment or alternatives.

My point is someone who shouts “att for any other pot” isn’t interested in doing it via RealmEye, which already exists for anyone who wants to use it. The types of offers, relative urgency, flexibility are all very different.

It might be an interesting academic exercise, but not IMHO at all useful.

#4

I think you’re taking a very narrow view on how it could be useful.
It’s not meant to be exactly the same as RealmEye trades, it’s meant to be a way to:
A) Filter for trades you’re interested in
B) See trade messages that you otherwise wouldn’t see
C) Attempt to standardize and make more digestible the wide variety of chat proposed trades

So, as an example of A – it is rather hellish trying to read/skim through 5 chats per second, but even more hellish looking only for any trades that are for your seasonal character, or for a particular item.
As an example of B, you could be on USEast, and see trades on USEast2 so you can jump over there and participate in said trade.
As an example of C, can use your example: “any pots except xxx”, the neural network can directly translate that into the pots that they most likely mean, and display that as a trade (using item icons rather whatever crude language and abbreviations), but also, you could potentially use that live info to set up filters and/or live alerts for trades you are actually interested in.

So, you may not like it, but it is far from useless.

#5

I’ve never had a problem. I generally ignore the scrolling chat though and wander around looking at text bubbles, rotating by 90° regularly to rearrange them. Alternately I announce the trade I want myself. If it’s a good trade (for someone else) I can see multiple replies in seconds, so clearly other people don’t have a problem reading the messages.

#6

While one could attempt to parse chat using text evaluation and/or regex to try and parse the many varying and crude abbreviations, reorderings, etc; a neural network would potentially be a lot better at it, with a lot fewer bugs, and any bugs could be slowly trained out – potentially on the fly. It could also be trained to detect ambiguity and use that to avoid trying to use ambiguous messages as trade proposals.

There are multiple possible models but, I’ve already started work using an RNN, but I may try out some others and see how they fare as well.

#7

This seems like a cool idea. Could this also be used to track an “average price” for items? As in, an average value that players are currently looking to pay for items, since realmeye’s trade offers feature could persist over days, but I imagine that real offers can vary by the minute (I know I tend to lower my prices if I spend too long without a reply to my current offer).

#8

Sure could. It’ll just be harder to do without cooperation from mreyeball

#9

I think it is a cool idea, and I can see how it would make the trade much more convenient.

However, I don’t think it would be something Deca would really be desperate to have for this game. Many other games have good trading mechanics, for example most have auction halls, and it would be very easy for players to trade that way. I think it’d be easy to make trading much more simpler if it is their intention.

Nevertheless, I would still like to see AI implemented into the game, in any way.

#10

a live bartering/auction hub where people put up offers and fulfill offers in real time would be nice
exists in most mmorpgs