4Game

Some outreach from the "toxic ranked battles".

I am very inspired when my roommate is watching LPL (China LOL league, kinda serious esports). During the live stream, there are occasional analyses of the situation along with the real-time winning probability of the two teams. Basically, various data (such as champion setup, gold difference, pieces of equipment, tower counts, positions and cooldowns, etc.) are taken into account and pushed into a neural network that outputs the winning probability.

And I instantly think about how this can supplement WOT. People are posting about selfishness in ranked battles and random queues, camping for damage, abandoning teammates for survival, etc. All these controversies actually have their counterparts in LOL. (like overfocusing on KDA or natural dmg imbalance between ADC/support) Yet in LOL esports, they seem to have a working solution and people don't often argue too much on who's the MVP and who's performing badly.

I imagine that if the postgame battle result (exp, chevrons) can be awarded based on a neural network, there will be fewer controversies on how players are performing. For example, we can calculate exp based on the winning probability influence of each action: killing shot, hitting an enemy, missing an enemy, bouncing shells, spotting, capping/decapping, or even long-range maneuvers/pushing and tanking/retreating. Estimation is that it will roughly work the same as the current "weighted sum" methods in simple scenarios like a 1v1, but will reveal and highlight more tactical value in a bigger picture. For example, an arty kill at 14-5 is apparently less valuable than an lt kill at 12-12. The same 400 dmg done at the first 2 minutes influence way more than that during the end of a 15-2 game. This may support a statistic that reveals more intel than MOE(combined damage) and win rates.

Just a vague idea. Friendly discussions are welcomed.