A Tutor Agent for MOBA Games

Victor do Nascimento Silva and Luiz Chaimowicz

Departament of Computer Science, Universidade Federal de Minas Gerais, Brazil

Abstract

Digital games have become a key player in the entertainment in-

dustry, attracting millions of new players each year. In spite of that,

novice players may have a hard time when playing certain types of

games, such as MOBAs and MMORPGs, due to their steep learn-

ing curves and not so friendly online communities. In this paper, we

present an approach to help novice players in MOBA games over-

Author’s Contact:

{vnsilva,chaimo}@dcc.ufmg.br

1 Introduction

Multiplayer Online Battle Arena (MOBA) games have become very

popular in recent years, reaching millions of users everyday. On a

first sight, MOBA games present a simple game concept: the player

controls a hero that should use its abilities to fight enemy heroes and

creeps, destroy structures and conquer an enemy base. Despite this

simple concept, this game genre have a complex gameplay and re-

quires a lot of knowledge that may not be so familiar to novice play-

Games often present a tutorial where the main gameplay is pre-

sented, aiming to teach newcomers how to play. Requiring the

player to complete tutorial matches or playing against bots is also

a very common practice, trying to assure that the player under-

stands the concepts presented. However, in MOBA games, there

are some features or roles that require the cooperation of more than

one player. To group players of similar skills in a team, these games

implement a matchmaking system [Mylak and Deja 2014], but, in

a cooperative scenario, this approach may not perform well. The

performance can be particularly poor in initial matches, as the com-

tips and support players, aiming at improving their experience and

gameplay while also benefiting the game by reducing the number

of dropouts. This tutor will replace one of the human players and

will go unnoticed by others, which will consider it a regular player.

Despite of the fact that the tutor supports all team, it is focused in

supporting a specific player, that it identifies as partner.

The rest of this paper is organized as follows: Section 2 presents

some definitions related to MOBA games and their player commu-

nity while in Section 3 we discuss some related works. Section 4

describes the mechanisms used for the tutor development and Sec-

2 MOBA Games

2.1 General Description

Multiplayer Online Battle Arena is a genre originated from Real

Time Strategy (RTS) games, in which players control an unit (Hero)

in a map, taking part in a battle to destroy an enemy base, like a cap-

ture the flag game. As the popularity of RTS games increased in the

last decade, players began to develop custom maps and features to

games, called MODs, using tools distributed with the game. Thus,

a custom map from Starcraft was the start point of MOBA games,

giving rise to Aeon of Strife. The popularity of this genre grew and

users from similar games began to develop similar maps. Defense

Dota 2, Heroes of the Storm, Heroes of Newerth and Strife. Data

from Riot Games shows that 67 millions of users plays matches

every month, with 27 millions of players daily and 7.5 millions of

player simultaneously [Games 2015].

MOBA gameplay generally consist of matches where ten players

compete, five in each team. At the beginning of a match, each

player must choose a hero, which he/she will be playing during

the entire match. The hero is a powerful unit with an unique spell

set that can be used to build complex game strategies and goals.

Controlling that hero, the player should join his teammates in a

makes the hero gain level, getting stronger and learning new abili-

ties.

2.2 Community

Real Time Strategy players are widely known by their expertise in

handling lots of information simultaneously and for also being very

“toxic”, raging at novice players [Kwak et al. 2015]. In Warcraft

III it is very common to kick players from matches when they are

identified as novice. Since DotA evolved from Warcraft III it is also

possible to kick players from games. When DotA became popu-

lar in online environments, the common sandbox were hold in a

Raging characteristics from RTS players seems to be connected

to modern MOBA games as most of them migrated from RTS to

MOBA. Valve approach to fight toxic behaviour in DotA 2 uses

a player report system. Riot also noticed toxic behavior amongst

LoL players, having initially created a system where players could

identify toxic behavior[Lin 2015].

As presented, MOBA community its not very friendly [Blackburn

and Kwak 2014], and that could imply in players feeling frustrated

more often. Some online games have similar behavior amongst

their players, and try to solve the problem by encouraging expe-

items, gold or titles in the community.

Csikszentmihalyi presents a model for identifying if players are in

flow [Csikszentmihalyi 1991]. Being in flow state means that play-

ers must not feel frustrated or bored, following in a way that the en-

tertainment is reached, having a good game environment and com-

munity can help players to be in flow state. However MOBA games

involves much more than just the difficulty presented by game me-

chanics. Some human factors, like toxic behavior, can drive players

to frustration. Playing in team requires cooperation and patience,

and that is what the matchmaking system from MOBA games tries

environment to watch or read tips. It is even harder to reach flow

when there is another player flaming at you by your mistakes or

inexperience.

3 Related Work

Tutorials are the main tool to interactively teach game features to

players in modern games. Normally, these teaching approaches are

simpler to apply since the game provides an environment for prac-

ticing what the player has learned [Silva and Nascimento 2012].

Although most game tutorials are successful to show simple game

features, it is not hard to find players that claim to have learned

MOBA game, and some of these games only show the mechanics

in a very high level. Thereby, most players claim that the tutori-

als are not sufficient to teach the basics of the game. Moreover, it

normally takes time to initially start playing a tutorial, then playing

with bots and finally Player versus Player (PvP). So, players gen-

rienced players playing in low level accounts, called “smurfs”. So

For developing the agent we extracted domain knowledge from ex-

pert players, in-game tutorials, videos and streaming from expert

players. We then divided this information in tactical and position-

accounts by visiting their profile. It is also possible to record in-

formation from the after match screen, where score information is

shown for each player who took part in the match. A brand new,

level 1 account was created assuring that LoL matchmaking would

select players with zero or low game knowledge.

For data collection, we tracked the player that partnered with the

tutor. We compared the results of three matches played prior and

three matches played right after the match played with the tutor.

Also, we ran 3 matches using only the supporting system and 3

matches with supporting and tip system enabled. All matches were

and how many times he/she died. This factor is broadly used to

analyse the gameplay and performance of players, both in amateur,

competitive or e-Sports scenario. The KDA factor function is pre-

sented in equation 1.

, D > 0

K + A, else

(1)

When comparing the matches using only the support system

(A,B,C) with support plus tip system (D,E,F) we can observe that

players which used the tip system had a larger performance im-

provement in matches played right after playing with the tutor. This

Figure 2: KDA performance of six players (A-F) before and after

players, except C, showed better than average performance, sup-

porting our hypothesis that the companion of a more knowledgeable

player improves performance. We can also observe that players D,

E, F showed much better performance when playing along our tutor,

showing that the tip and support system together is helpful to new

players. Overall, we observe that all players, except by C, showed

performance improvements after playing along the tutor , showing

that the tutor may help them to improve their gameplay. Player C

be considered normal, as these players are learning how to play.

Results shown that players B, C and E reduced the standard devia-

tion, meaning they became more stable in their gameplay. On the

other hand, we observed that players A, D and F showed higher

standard deviations after playing along the tutor. Standard devia-

tion, in our data collection, can be influenced by many factors like

the performance increase, player experience, connection problems,

and others, that means it should not be taken as a conclusive evalu-

ation factor in this study.

Finally, League of Legends provides a chat room in game and after

it. It also provides an honor system, in which partners can honor

teammates for being helpful, friendly or showing teamwork. More-

over, LoL lets users add others as friends, so they can invite others

to play matches together. After the three matches with tip system

Answer Number of answers Percentage

Good frequency 4 66.7%

Little low frequency 1 16.6%

Low frequency 1 16.6%

5.2 Experiment 2: Survey and play with controlled

For evaluation of our approach we performed some experiments

with players. We invited random players to join our agent in two

matches. In the first one the player would play alongside the agent

without the tip system. In the second match, the player would play

We have a set of six players, all male with ages ranged from 18

to 24 years old. In the first survey we collected data from their

experience with MOBA and League of Legends. The experience of

these players could reveal how well they could evaluate our system.

Also, we have asked players about their experience with MOBAs

others than LoL, evaluating the overall expertise of these players.

All players were familiar with MOBA, three of them were ad-

vanced, two intermediate and one beginner. We notice that the ma-

jority or players are advanced, and we expect these players could

provide us valuable feedback about the tutor. In addition, all of

tutorial of median to poor value. We had two players that did not

do the tutorial, two that attributed a score of four and two that gave

a score of five to the LoL tutorial. In addition we asked if the player

considers that the tutorial is sufficient for teaching a new player

the basic concepts of the game obtaining. We also asked questions

about the hero, the game and player experience, however we will

not include all details in this work due to paper size.

After the matches played alongside our agent, we asked the player

to answer a survey about his experience with the agent alone and

the agent plus the tip system. The survey revealed that none of

actions that the agent performed during the match, classifying him

as robotic or human. Four of these players considered the agent

more robotic than human, one considered it very robotic and one

considered it nor robotic nor human.

6 Conclusions and Future Work

In this paper we addressed the problem of helping novice players

with the features and mechanics of MOBA games aiming to im-

prove their entertainment. For this, we develop an Artificial In-

telligence driven agent that plays along players guiding and tip-

and experience with the tutor. They were also invited to play two

matches along the agent. Results show promising results towards

helping new players, as these players evaluate that the agent per-

formed well as support. They also reported that the frequency of

the tips are well balanced in most cases.

For future work, we intend to perform more experiments, both qual-

itative and quantitative. Also, a deeper analysis can be done by

tracking players’ progression using the tutor studying the period

needed to learn game features and agent lifetime. Another analy-

sis that can be performed is the gameplay improvement of the entire