"GenLadder SUS & GEQ Questionnaire"

"Recent advances in Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs), have enabled dynamic and adaptive content generation in interactive applications such as digital games. However, the use of GenAI in logic-constrained puzzle games introduces significant challenges related to correctness, solvability, and consistent difficulty control. This paper presents GenLadder, an AI-driven word ladder game that integrates LLM-based puzzle generation with a deterministic, graph-theoretic validation framework to ensure linguistic validity and guaranteed solvability. Personalized word ladder puzzles are generated in real time based on player proficiency and selected difficulty levels, while a multi-stage validation pipeline filters invalid content and verifies solvability using breadth-first search (BFS) over length-specific word graphs. The BFS solver guarantees optimal transformation paths and enables objective difficulty classification based on minimal path length.To evaluate usability and player experience, a quantitative user study was conducted with 31 participants who played the GenLadder mobile application and completed standardized questionnaires. Usability was assessed using the System Usability Scale (SUS), while player experience was evaluated using the Game Experience Questionnaire (GEQ). Results indicate that GenLadder achieved high perceived usability, exceeding established SUS benchmarks, and demonstrated strong player engagement, particularly in the dimensions of challenge, immersion, and flow. These findings suggest that combining GenAI-driven content generation with formal algorithmic validation can produce scalable, reliable, and engaging puzzle games. Beyond the word ladder domain, the proposed architecture provides a generalizable blueprint for safely integrating generative AI into interactive systems that require strict logical constraints."

近年来，生成式人工智能（Generative Artificial Intelligence，GenAI），尤其是大语言模型（Large Language Models，LLMs）的研究进展，为数字游戏等交互应用带来了动态自适应的内容生成能力。然而，在逻辑约束型益智游戏中使用GenAI，会带来与正确性、可解性及难度一致性控制相关的诸多严峻挑战。本文提出了GenLadder——一款由人工智能驱动的单词接龙游戏，它将基于大语言模型的谜题生成与确定性图论验证框架相结合，以确保语言合规性与可解性保障。该系统可根据玩家熟练度与所选难度等级实时生成个性化单词接龙谜题，同时通过多阶段验证流程过滤无效内容，并基于针对词长的单词图使用广度优先搜索（breadth-first search，BFS）验证谜题可解性。该BFS求解器可确保生成最优转换路径，并基于最小路径长度实现客观的难度分级。为评估系统可用性与玩家体验，研究团队招募31名参与者开展了定量用户研究，所有参与者均体验了GenLadder移动端应用并完成了标准化问卷。可用性评估采用系统可用性量表（System Usability Scale，SUS），玩家体验评估则采用游戏体验问卷（Game Experience Questionnaire，GEQ）。研究结果显示，GenLadder获得了较高的主观可用性评分，超过了SUS既定基准，且展现出出色的玩家参与度，尤其在挑战性、沉浸感与心流体验维度表现突出。上述研究结果表明，将GenAI驱动的内容生成与形式化算法验证相结合，可打造出可扩展、可靠且极具吸引力的益智游戏。除单词接龙领域外，本文提出的架构还为将生成式人工智能安全集成至需严格逻辑约束的交互系统提供了可推广的蓝图。