PaperlayerView exact revision on arXiv

Source-linked AI summary

LeagueBot: A Voice LLM Companion of Cognitive and Emotional Support for Novice Players in Competitive Games

Jungmin Lee, Inhee Cho, Youngjae Yoo

arXiv:2602.01213v1cs.HC

TL;DR

Novice players face steep learning curves and social pressure in competitive games, while evidence on voice-based LLM support in competitive play remains limited. This study developed and evaluated LeagueBot, a voice chatbot for cognitive and emotional support during League of Legends gameplay. The authors report that the system enhanced player experience and identify a single-session evaluation as a limit on assessing longer-term outcomes.

  • Problem

    Research directly examining voice-based LLM support in competitive gameplay remains limited, despite novice players’ steep learning curves and social pressures.

    Source: §1 · pp. 2–3

  • Method

    The study developed a voice-based LLM agent for novice League of Legends players and evaluated it using a within-subjects experiment and qualitative interviews.

    Source: §1 · p. 3, §2.4 · p. 6

  • Results

    The study reports that integrating LLM-based voice agents into competitive games can enhance player experience and identifies design considerations for fast-paced, high-pressure settings.

    Source: §1 · p. 3

  • Takeaways & Limitations

    LeagueBot was designed to combine access to complex game knowledge with emotional support intended to help alleviate novice players’ psychological pressure.

    Source: §3.2.1 · p. 6

    The study involved only a single gameplay session, so it could not assess whether continued chatbot use sustains or improves performance.

    Source: §6.3 · p. 25

Abstract

from arXiv · show

Competitive games pose steep learning curves and strong social pressures, often discouraging novice players and limiting sustained engagement. To address these challenges, this study introduces LeagueBot, a large language model-based voice chatbot designed to provide both informational and emotional support during live gameplay in league of legends, one of the most competitive multiplayer online battle arena games. In a within-subjects experiment with 33 novice players, LeagueBot was found to reduce cognitive challenge, performative challenge, and perceived tension. Qualitative analysis further identified three themes: enhanced access to game information, relief from cognitive burden, and practical limitations. Participants noted that LeagueBot offered context-appropriate guidance and emotional support, helping ease the steep learning curve and psychological pressures of competitive gaming. Together, these findings underscore the potential of voice-based LLM companions to assist novice players in competitive environments and highlight their broader applicability for real-time support in other high-pressure contexts.

1 INTRODUCTION

Competitive games, particularly MOBAs, combine steep learning curves with real-time demands and social pressure that can overwhelm novice players. The study addresses limited research on voice-based LLM support in competitive gameplay by developing and evaluating a prototype for cognitive and emotional support.

1 INTRODUCTION

  • MOBAs require extensive game knowledge and complex strategies, while frequent updates can intensify their learning curve and overwhelm beginners.Team mistakes and toxic or uncooperative community cultures can compound social pressure and contribute to dropout.Source: §1 · p. 2
  • Prior competitive-game support tools have mainly targeted technical and strategic performance, while psychological pressure and emotional support have received comparatively less attention.Existing LLM game-companion research has largely examined slower-paced, mostly text-based environments.Source: §1 · p. 2
  • Competitive games require fast decisions and complex controls, making text-based assistance insufficient for their pace.Voice interaction has been effective in similarly demanding contexts such as driving.Source: §1 · p. 2
  • The study asks what cognitive and emotional experiences players encounter, what support a voice agent should provide, and what benefits and limitations it may have.It develops a prototype and evaluates it in competitive gameplay using a within-subjects, mixed-methods approach.Source: §1 · p. 3
  • The authors report that integrating an LLM voice agent can enhance player experience and identify design considerations for fast-paced, high-pressure environments.The evaluation combined quantitative experiments with qualitative interviews.Source: §1 · p. 3

2 RELATED WORKS

The related work describes competitive play as both demanding and rewarding, with novices facing cognitive, performance, and social pressures. Existing support tools and LLM companions offer relevant precedents, but voice-based LLM support in dynamic competitive games remains little studied.

2.1 Player experience in competitive games

  • Competitive games can offer achievement and satisfaction, yet steep difficulty and skill gaps can make play especially challenging for beginners.MOBAs are described as producing greater frustration and challenge than many other genres.Source: §2.1 · pp. 3–4

2.2 Psychological burden of competitive games

  • Team competition and toxic behavior can intensify novice frustration and hostility, while players use individual and social strategies to manage psychological pressure.Prior work identifies teammate conflict as a major stressor and friend-based play as a source of support.Source: §2.1 · pp. 3–4, §2.2 · p. 4

2.3 Computational tools for supporting competitive games

  • Competitive-game tools provide contextual feedback and assistance, but most remain narrowly focused on technical and strategic functions rather than emotional support.Prior work describes principles including situational awareness, reasoning explanations, and adaptation to skill level.Source: §2.3 · p. 5

2.4 Large language models as game companions

  • LLM companions have supported guidance, creativity, and emotional engagement in games, but prior work mostly studied slower-paced settings where text interaction suffices.Fast-paced competitive play motivates voice-based support, yet its effectiveness, acceptance, and feasibility remain little examined.Source: §2.4 · pp. 5–6

3 DEVELOPMENT OF LEAGUEBOT

LeagueBot was designed for novice League of Legends players, a population facing a steep learning curve and social pressure in a complex, fast-changing game. The development process combined real-time voice support with pilot feedback that shaped the system’s interactions and scope.

3.1 Selection of a competitive game

  • League of Legends was selected as a demanding setting because players must master many champions, items, strategies, and frequent balance changes.Newcomers also face psychological pressure and teammate conflict that can make adaptation difficult.Source: §3.1 · p. 6

3.2.1 Design goals

  • LeagueBot’s design goals were real-time adaptation, efficient access to complex game knowledge, and emotional support for novice players.The intended companion offers comfort and empathy while helping alleviate psychological pressure.Source: §3.2.1 · p. 6

3.2.2 Implementation of LLM-based voice chatbot

  • The chatbot used ElevenLabs’ conversational AI and JavaScript SDK in an Electron desktop application, with GPT-4.1 as its LLM.A Korean-supported male voice was selected for the primary user base.Source: §3.2.2 · p. 7

3.2.3 Chatbot persona

  • Prompt engineering shaped LeagueBot as a supportive teammate, with friendly, cooperative interactions intended for immediate, context-sensitive in-game guidance.The persona drew on a design framework and esports coaching methods.Source: §3.2.3 · p. 7

3.2.4 Game information and context awareness

  • LeagueBot used contextual updates and on-demand tools alongside live game and player data to tailor strategic and personalized advice.Game data informed advice about match flow and objectives; player data supported recommendations on skills, items, farming, or combat.Source: §3.2.4 · p. 7

3.3 Pilot Test and Final Design

  • A six-person pilot used one full game and an immediate interview to assess LeagueBot’s design and identify live-game issues.The pilot included players with varied experience, and its feedback was used to refine the final design shown in Figure 1.Source: §3.3.1 · p. 8, Figure 1 · p. 8

3.3.2 Feedback and implementation

  • The pilot found generally positive responses to information delivery and emotional responses, while identifying areas for improvement.Participants objected to unsolicited speech during concentration and preferred concise, direct, front-loaded guidance; proactive speech was disabled and prompts were revised.Source: §3.3.2 · p. 9
  • The final revisions varied reassurance, added voice choices, and extended LeagueBot support into champion selection.The temperature was raised from 0 to 0.5 to vary emotional responses, and contextual updates refreshed team and opponent information during selection.Source: §3.3.2 · p. 9

4 METHOD

The study compared LeagueBot-assisted play with unassisted play in a randomized within-subjects design, then used interviews to examine participants’ experiences. Participants were recruited as novice League of Legends players under a specified eligibility criterion.

4.1 Study design

  • Each participant played both with and without LeagueBot in randomized order, followed by an interview about their experience.The study combined quantitative experiments and qualitative interviews, allowed usual game configurations and supplementary guides, and disabled in-game text chat.Source: §4.1 · p. 10

4.2 Participants

  • Novice status was defined as being below level 30 or having played no more than 20 ranked matches after reaching level 30.The authors used limited ranked experience as an indicator because ranked play involves greater competitiveness and complexity.Source: §4.2 · p. 10
  • Thirty-four participants met the recruitment criteria, and one was excluded from final analysis after being identified as having sufficient ranked experience.The resulting participant pool included varied play frequency, preferred modes, and competitive orientation.Source: §4.2 · pp. 10–11

Appendix 2.

The study compared LeagueBot with ordinary play across two matches, measuring challenge, experience, usability, and interview responses. Gameplay and interaction records complemented paired-condition quantitative tests and thematic analysis.

4.3 Procedure

  • Participants played one 20–30-minute match per session in each condition; LeagueBot was available throughout the experimental match and used real-time game data for context-sensitive advice.Control players could consult online resources; LeagueBot users could ask about champion abilities, items, and strategy, including during warm-up.Source: §4.3 · p. 11

4.4.1 Measures

  • Measures covered three CORGIS challenge dimensions, IMI enjoyment and tension, and LeagueBot usability using the 10-item System Usability Scale.The emotional CORGIS dimension was excluded as less relevant to competition-focused LoL; SUS scores range from 0 to 100.Source: §4.4.1 · pp. 11–12

4.4.2 Interviews

  • Semi-structured interviews asked about gameplay with and without LeagueBot, its role, comparison with a human teammate, information sources, and improvement suggestions.Interviews lasted approximately 20–30 minutes and were recorded and transcribed with consent.Source: §4.4.2 · p. 12

4.4.3 Game data

  • Timestamped chatbot logs and screen recordings captured when and how LeagueBot interactions unfolded during actual matches.These data were analyzed alongside interview results.Source: §4.4.3 · p. 12

4.5 Data analysis

  • Paired t-tests compared condition means, while thematic analysis used iterative author review and workshops to agree on final interview themes.The first author conducted initial coding after transcripts were manually corrected.Source: §4.5 · pp. 12–13

5 Results

LeagueBot was associated with lower challenge ratings and tension, while gameplay outcomes showed no significant condition differences. The study also reports measurement checks and descriptive statistics for its sample.

5.1 Measurement validation

  • Five constructs showed adequate convergent validity and reliability after low-loading items were removed; retained item loadings exceeded .6, AVE exceeded .5, CR exceeded .6, and Cronbach’s α exceeded .7.The removed items comprised four cognitive-challenge, five decision-challenge, and one enjoyment item.Source: §5.1 · p. 13

5.2 Descriptive statistics

  • The study reported descriptive statistics for dependent variables, game outcome data, and chatbot usage data; LeagueBot’s SUS score was 76.7, indicating above-average usability.Mean KDA was 3.47 (SD = 3.80) in the chatbot condition versus 2.54 (SD = 2.72) in the non-chatbot condition.Source: §5.2 · pp. 14–15

5.3 Results of gameplay

  • 45% chatbot versus 55% non-chatbot win rates did not differ significantly; game duration and KDA differences were also nonsignificant.The authors report no substantial influence on win rates, duration, or combat performance.Source: §5.3 · p. 15

5.4 Effectiveness of LeagueBot

LeagueBot users reported lower cognitive and performative challenge and tension than in the no-chatbot condition, while enjoyment and decision-making challenge did not differ. The results suggest effects on some, but not all, measured aspects of play.

5.4 Effectiveness of LeagueBot

  • Decision-making challenge and enjoyment showed no significant condition differences, with reported p-values of .862 and .724, respectively.The corresponding effect sizes were d = 0,03 for decision-making challenge and d = -0.06 for enjoyment.Source: §5.4 · p. 16
  • 0.68-point cognitive challenge, 0.59-point performative challenge, and 0.47-point tension reductions favored LeagueBot; decision-making challenge and enjoyment were unchanged.The reported differences were significant for cognitive challenge (p = .002), performative challenge (p = .003), and tension (p = .043), but not for decision-making challenge or enjoyment.Source: Figure 2 · p. 16, §5.4 · p. 16

5.5 Results of thematic analysis

The thematic analysis identified three themes: advantages of acquiring game information, burden alleviation, and practical pros and cons. Participants described context-aware information and emotional support alongside limitations in live use.

5.5.1 Advantages of acquiring game information

  • Voice access provided immediate in-game information without leaving the screen, though simultaneous talking and play sometimes disrupted concentration or immersion.Some participants said they lacked time to search during matches; others found listening while playing difficult.Source: §5.5.1 · p. 17
  • LeagueBot offered situation-specific reasoning and advice on concrete actions, extending beyond statistics to help players understand and respond to game situations.Participants asked about ward placement and opposing-team composition; an example shows P5 using advice to join a mid-lane teamfight.Source: §5.5.1 · pp. 17–18, Figure 3 · p. 18
  • Participants said the chatbot let them ask basic questions without burdening or being judged by friends, lowering a psychological barrier to learning.They described expecting supportive answers rather than criticism.Source: §5.5.1 · pp. 18–19

5.5.2 Burden alleviation through the Chatbot

  • By enabling immediate questions and responsive feedback, LeagueBot eased novices’ anxiety about limited knowledge and uncertainty in their decisions.Participants also said it reduced worries about learning game rules and strategies.Source: §5.5.2 · p. 19
  • Encouragement and consolation helped participants manage competitive pressure and frustration, while some valued the chatbot as a companion during solo play.One participant described feeling comforted after a mistake; another called the chatbot a valuable in-game partner.Source: §5.5.2 · pp. 20–21

5.5.3 Pros and cons for the real-world implementations

  • Participants reported that real-time guidance reduced cognitive load and fatigue, and that speech recognition handled game jargon in authentic gameplay.Some described the speech recognition and response speed as faster than expected.Source: §5.5.3 · p. 21
  • Participants also cited text-only item names, verbosity, and slow-to-arrive conclusions as practical obstacles to quick decisions.The lack of item images was inconvenient for players unfamiliar with item names.Source: §5.5.3 · pp. 21–22
  • Hallucinated recommendations and response delays reduced effectiveness; delays were especially disruptive when matches required immediate decisions.Participants reported inaccurate details or nonexistent items and sometimes avoided asking during fights.Source: §5.5.3 · p. 22

6 Discussion

LeagueBot reduced novices’ cognitive and performative challenges and tension through timely, context-appropriate information and emotional support. The discussion considers its theoretical implications, design directions, and limits.

6 Discussion

  • LeagueBot’s timely, context-appropriate support reduced cognitive burdens and performative challenges for novice players, including through reassurance and empathy.It provided information without making decisions for players.Source: §6 · pp. 22–23
  • Players’ tension decreased, while enjoyment showed no considerable difference; reassurance and companionship were possible contributors to a more relaxed experience.Interview findings linked reduced tension to help with unfamiliar rules and strategies, alongside emotional support.Source: §6 · p. 23

6.1 Theoretical implications

  • Interactive dialogue can provide context-aware, reasoned guidance and let novices ask questions freely, potentially supporting learning and skill development.The authors contrast this with tools requiring players to leave the game screen.Source: §6.1 · pp. 23–24
  • The study extends evidence for LLM-based emotional support from slower-paced settings to competitive MOBA play, where such support had been little investigated.The authors also suggest possible relevance to FPS games and esports players’ performance-related stress.Source: §6.1 · pp. 23–24

6.2 Design implications

  • Design should enrich gameplay context beyond LoL API data and adapt explanation detail and vocal style to the immediacy of the situation.Detailed reasoning may suit calm moments, while concise, directive guidance may better fit critical moments.Source: §6.2 · pp. 24–25
  • Inaccurate responses and response latency remain a design trade-off: retrieval can add latency, while fine-tuning is resource-intensive and less suited to frequent updates.The authors propose combining fine-tuning on core information with minimal, context-relevant updates via prompting.Source: §6.2 · p. 25
  • Some participants preferred friends’ guidance and social conversation, motivating longer-term interaction and personalization rather than informational support alone.The authors suggest conversational histories could help tailor advice to players’ styles, champions, and recurring mistakes.Source: §6.2 · p. 25

6.3 Limitations and future works

  • Generalizability is constrained by a relatively small sample, one gameplay session, and a focus on LoL and the MOBA genre.The study therefore cannot establish whether learning or performance effects persist with continued use or transfer to other game contexts.Source: §6.3 · p. 25

7 CONCLUSION

The study evaluated LeagueBot, an LLM-based voice chatbot, in competitive gameplay using mixed methods. It found reduced novice players’ cognitive and performative challenges and tension, alongside practical challenges for the system.

7 CONCLUSION

  • LeagueBot’s timely, context-aware responses mitigated novice players’ cognitive and performative challenges while reducing tension.The study used a mixed-method approach in competitive gaming contexts.Source: §7 · p. 25
  • Qualitative findings indicate that LeagueBot eased the burden of asking questions and provided socio-emotional support, with assistance participants regarded as more effective than conventional tools or peer interactions.The conclusion also associates this support with facilitating learning and enhancing immersion.Source: §7 · p. 26
  • The conclusion identifies latency, speech-recognition accuracy, and dynamically calibrating information as technical and design challenges.It suggests sustained interaction as a path toward a companion-like system supporting immersion and learning.Source: §7 · p. 26

A.1 LeagueBot Prompt

The prompt configures LeagueBot as a casual, supportive League of Legends voice companion. It specifies how the bot uses game context, responds to gameplay and emotional cues, and keeps its speech brief.

A.1 LeagueBot Prompt

  • LeagueBot is framed as a Korean-speaking companion for casual League of Legends players, focused on high-level gameplay reflection and motivation.Its intended voice is friendly and supportive, like a teammate rather than a coach.Source: §A.1 · pp. 29, 30, 32
  • Responses should be casual, natural Korean and usually one or two sentences, with advice stated first and briefly explained.The prompt cautions against over-explaining during high-pressure gameplay.Source: §A.1 · pp. 29, 31, 32
  • During champion select, the bot receives the user’s champion and role, then updates team and opponent picks and bans as the draft progresses.The prompt directs it to answer based on available context rather than request game information unnecessarily.Source: §A.1 · pp. 29–30
  • Once a match begins, LeagueBot uses game and user-state tools for match details, including score, game time, champion, items, gold, and lane role.The prompt asks for relevant, timely, friendly responses using this data.Source: §A.1 · p. 30
  • The bot’s role was to help users reflect on high-level gameplay decisions and stay motivated, offering context-specific strategic guidance and emotional comfort when things went wrong.Examples include suggesting vision and another fight after a dragon-fight loss, or farming CS to catch up after an early lane death.Source: §A.1 · pp. 29–31

A.2 Participants

The appendix lists participant-level profiles with demographic, League of Legends experience, play, competitiveness, and LLM-use fields.

A.2 Participants

  • The participant table records gender, age, play frequency, LoL level and play period, competitiveness, main mode, and LLM usage.These fields appear in the table header.Source: §A.2 · pp. 32–33
  • Listed profiles include participants with different ages, LoL levels, play histories, and reported play frequencies.For example, the entries include participants aged 20 and 27, and both infrequent and weekly play.Source: §A.2 · pp. 32–33
  • The profiles also record different main modes and LLM-use frequencies, including ARAM and Normal Game players and both daily users and never-users.These are examples from the listed participant rows.Source: §A.2 · p. 33
Loading 2602.01213v1…