The effect of AI-based inspiration on human design ideation
基于人工智能灵感对人类设计构思的影响

2.1. Computational co-creative systems
2.1. 计算协作系统

The study of computational co-creative systems is one of the growing fields in computational creativity that involves human users collaborating with an AI agent to design creative artifacts. Co-creativity is a collaboration in which multiple parties collaboratively and synthetically contribute to the creative process in a blended manner (Mamykina et al., Citation2002).
计算协作系统的研究是计算创造力中一个不断发展的领域，涉及人类用户与人工智能代理合作设计创意作品。共创是多方协作并综合性地为创意过程做出贡献的合作方式（Mamykina 等，2002 年）。

Co-creative systems have been applied in different creative domains such as art, music, dance, drawing, and game design. The computational agent in a co-creative system can directly perform actions on a shared artifact, whereas others provide suggestions to inspire users for generating novel ideas. How a co-creative AI agent contributes to the creative process is a factor that distinguishes different co-creative systems.
共创系统已应用于不同的创意领域，如艺术、音乐、舞蹈、绘画和游戏设计。共创系统中的计算代理可以直接在共享的作品上执行动作，而其他人则提供建议，以激发用户产生新颖的想法。共创人工智能代理如何促进创意过程是区分不同共创系统的因素。

One of the co-creative interaction paradigms in design is the turn-taking action between a user and an AI agent in a shared artifact. Drawing Apprentice (Davis, Hsiao, et al., Citation2015) is a web-based co-creative drawing system that analyzes the user’s sketch and responds to the user’s sketch. In the Drawing Apprentice, the user starts drawing a sketch on the canvas, then the AI agent generates and adds a sketch based on the users’ sketch. DuetDraw (Oh et al., Citation2018) is another co-creative drawing system that enables users and the AI agent to draw pictures collaboratively. DuetDraw helps users perform drawing tasks, such as completing the rest of the object that the user was drawing, drawing the same object in a different style, suggesting an object that matches the picture, finding an empty space on the canvas, and automatically colorizing the sketches. Cobbie (Lin et al., Citation2020) is a mobile robot embedded with recurrent neural network (RNN)-based co-creative methods and mobile drawing system to support early-stage ideation. Cobbie provides inspirational sketches under the command of the designer.
设计中的共创互动范式之一是用户与 AI 代理在共享工件中的轮流行动。《绘画学徒》（Davis，Hsiao 等，2015）是一个基于网络的共创绘画系统，分析用户的草图并响应用户的草图。在《绘画学徒》中，用户在画布上开始绘制草图，然后 AI 代理根据用户的草图生成并添加草图。《双人绘画》（Oh 等，2018）是另一个共创绘画系统，使用户和 AI 代理能够协作绘制图片。《双人绘画》帮助用户执行绘图任务，例如完成用户正在绘制的对象的其余部分，以不同风格绘制相同的对象，建议与图片匹配的对象，找到画布上的空白空间，并自动给草图上色。《Cobbie》（Lin 等，2020）是一个嵌入了基于递归神经网络（RNN）的共创方法和移动绘图系统的移动机器人，用于支持早期构思。《Cobbie》在设计师的指令下提供灵感草图。

While the co-creative interaction paradigms above show the examples that an AI agent is directly involved in a creative activity by performing the same type of action with a user, another co-creative interaction paradigm provides suggestions to the user. The Sentient Sketchbook (Yannakakis et al., Citation2014) and 3Buddy (Lucas & Martinho, Citation2017) are co-creative systems for game level design. In both systems, the AI agent provides feedback and additional ideas to develop the game design. These systems use a turn-taking interaction but provide suggestions to the human designer rather than creating game levels directly.
以上的共创互动范式展示了 AI 代理直接参与创意活动的示例，通过与用户执行相同类型的动作。另一个共创互动范式为用户提供建议。《有感知的素描本》（Yannakakis 等，2014）和 3Buddy（Lucas＆Martinho，2017）是用于游戏关卡设计的共创系统。在这两个系统中，AI 代理提供反馈和额外的想法来发展游戏设计。这些系统采用轮流互动，但提供建议给人类设计师，而不是直接创建游戏关卡。

Some recent studies investigated the role of AI and the impact of AI on ideation in human-AI collaboration. Liao et al. (Citation2020) presented three potential roles of AI-based inspirations in ideation that are closely related to the interaction paradigm providing suggestions described above: AI as representation creation (providing inspirations by suggesting texts or images), AI as an empathy trigger (supporting the designer’s descriptive thinking), and AI as engagement (helping the designer avoid fossilization and perform typical design actions). Figoli et al. (Citation2022) claimed that the role of AI in human-AI collaboration relies on the capability of AI (i.e., AI as a teammate when AI performance is better than human performance and AI as external stimuli when AI performance is worse than human performance). While the roles of AI emphasize the positive impact of AI in human-AI collaboration, Pandya et al. (Citation2019) and Zhang et al. (Citation2021) showed that human-AI collaboration can produce better outcomes when the AI contributes a better performance than the human. In other words, it is likely to produce poorer outcomes if the AI performs the same or worse than the human. These studies indicate that the roles and effects of AI in human-AI collaboration need further exploration.
一些最近的研究调查了人工智能在人工智能协作中的作用以及对构思的影响。廖等人（2020）提出了人工智能基于灵感在构思中的三种潜在角色，这些角色与上述提供建议的交互范式密切相关：人工智能作为表征创造（通过建议文本或图像提供灵感）、人工智能作为共情触发器（支持设计师的描述性思维）以及人工智能作为参与（帮助设计师避免僵化并执行典型的设计行为）。Figoli 等人（2022）声称人工智能在人工智能协作中的角色取决于人工智能的能力（即当人工智能表现优于人类表现时，人工智能作为队友，当人工智能表现不如人类时，人工智能作为外部刺激）。尽管人工智能的角色强调了人工智能在人工智能协作中的积极影响，但 Pandya 等人（2019）和 Zhang 等人（2021）表明，当人工智能的表现优于人类时，人工智能协作可以产生更好的结果。换句话说，如果人工智能的表现与人类相同或不如人类，可能会产生较差的结果。 这些研究表明，人工智能在人工智能协作中的角色和影响需要进一步探讨。

A co-creative system that is closely related to our co-creative system is the Creative Sketching Partner (CSP) (Davis, Siddiqui, et al., Citation2019; Karimi, Rezwana, et al., Citation2020). The CSP presents sketches of varying visual and conceptual similarity based on the designer’s task and sketch. Users can control the parameters of the algorithm by specifying the amount of visual and conceptual similarity. Karimi, Rezwana, et al. (Citation2020) focuses on identifying relationships between the AI model of visual and conceptual similarity and three types of design creativity: combinatorial, exploratory, and transformational. The findings suggest that inspiration related to conceptual similarity is more associated with transformational creativity and inspiration related to visual similarity is more associated with combinatorial creativity. In contrast, this paper focuses on the cognitive process of ideation rather than the creativity of the outcome. A comparison of the Comparison of Creative Sketching Partner (Davis, Siddiqui, et al., Citation2019; Karimi, Rezwana, et al., Citation2020) and the Collaborative Ideation Partner (this paper) is shown in
一个与我们的共创系统密切相关的共创系统是创意素描伙伴（CSP）（Davis，Siddiqui 等，2019 年；Karimi，Rezwana 等，2020 年）。CSP 根据设计师的任务和素描呈现不同视觉和概念相似性的素描。用户可以通过指定视觉和概念相似性的数量来控制算法的参数。Karimi，Rezwana 等人（2020 年）专注于识别视觉和概念相似性 AI 模型与三种设计创造力（组合、探索和转化）之间的关系。研究结果表明，与概念相似性相关的灵感更与转化创造力相关，而与视觉相似性相关的灵感更与组合创造力相关。相比之下，本文侧重于构思的认知过程而非结果的创造力。创意素描伙伴（Davis，Siddiqui 等，2019 年；Karimi，Rezwana 等，2020 年）与协作构思伙伴（本文）的比较如下所示。Table 1 表 1.

2.2. Design ideation 2.2. 设计构思

Creativity is strongly associated with ideation. Creativity is defined as a process leading to a creative outcome, novel and useful products, and the ability to generate such work (Amabile, Citation1982; Jagtap, Citation2019; Oman et al., Citation2013; Weisberg, Citation1993). Ideation is a creative process where designers generate, develop, and communicate new ideas. Ideation in design can lead to innovative design solutions through generating diverse concepts (Akin, Citation1990; Atman et al., Citation1999; Brophy, Citation2001; Cross, Citation2001; Daly, Yilmaz, et al., Citation2012; Liu et al., Citation2003). The goal of the design is to develop useful and innovative solutions and design ideation that allows designers to explore different areas of the design solution space (Daly, Christian, et al., Citation2011; Newell & Simon, Citation1972). The design process is an evolution of different kinds of representations (Goel & Pirolli, Citation1992). In the design process, designers exteriorize and visualize their design intentions and communicate with external visualizations to interact with their internal mental images (Dorta, Citation2008).
创造力与构思紧密相关。创造力被定义为导致创造性成果、新颖且有用的产品以及产生这类作品的过程（Amabile, 1982; Jagtap, 2019; Oman 等，2013; Weisberg, 1993）。构思是设计师生成、发展和传达新想法的创造过程。设计中的构思可以通过生成多样化的概念导致创新的设计解决方案（Akin, 1990; Atman 等，1999; Brophy, 2001; Cross, 2001; Daly, Yilmaz 等，2012; Liu 等，2003）。设计的目标是开发有用和创新的解决方案，设计构思使设计师能够探索设计解决方案空间的不同领域（Daly, Christian 等，2011; Newell & Simon, 1972）。设计过程是不同类型表征的演变（Goel & Pirolli, 1992）。在设计过程中，设计师外化和可视化他们的设计意图，并通过外部可视化与内部心理形象互动交流（Dorta, 2008）。

Analogy is an ideation method relevant to this study. Analogy in design facilitates creative thinking by suggesting potential solutions (Jia et al., Citation2020). Analogical reasoning is an inference method in design cognition that leads to unexpected discoveries (Gero & Maher, Citation1991). Design-by-Analogy (DbA) is a design tool that provides inspiration for innovative design solutions (Christensen & Schunn, Citation2007; Fu et al., Citation2013; Goel, Citation1997; Jia et al., Citation2020). Inspirations in Design-by-Analogy (DbA) are achieved by transferring a design problem (source) to a solution (target) in another domain (Moreno et al., Citation2014). The association between a source design and a target design can be either semantic (conceptual) characteristics or visual (structural) representations. Semantic and visual stimuli can thus serve as a basis for developing computational systems that support design ideation.
类比是与本研究相关的一种构思方法。设计中的类比通过提出潜在解决方案来促进创造性思维（贾等，2020 年）。类比推理是设计认知中的一种推理方法，可以导致意想不到的发现（杰罗和马赫，1991 年）。类比设计（DbA）是一种设计工具，为创新设计解决方案提供灵感（克里斯滕森和舒恩，2007 年；傅等，2013 年；戈尔，1997 年；贾等，2020 年）。类比设计中的灵感是通过将设计问题（源）转移到另一个领域的解决方案（目标）来实现的（莫雷诺等，2014 年）。源设计与目标设计之间的关联可以是语义（概念）特征或视觉（结构）表示。因此，语义和视觉刺激可以作为开发支持设计构思的计算系统的基础。

Many computational approaches and tools using analogy have been developed based on semantic distance (conceptual similarity) to support ideation. Zuo et al. (Citation2022) presented a semantic network based on Wikipedia called WikiLink, a design ideation tool for design innovation. To provide explorations of knowledge for design innovation, WikiLink employs a weight that fuses the statistical and semantic relationships, which captures the implicit connection between concepts. Kang et al. (Citation2021) presented a supporting tool, MetaMap, which inspires visual metaphor ideation through multi-dimensional example-based exploration. To facilitate the ideation process, MetaMap provides example-based exploration in semantics, color, and shape dimensions with sample images based on keyword association and color filtering. J. Luo, S. Sarica, and K. L. Wood (Citation2021) proposed a knowledge-based expert system to provide computationally guided exploration and exploitation of design stimuli taken from the total patent database at the concept, document, and field levels simultaneously. Knowledge distance guides the network-based exploration and retrieval of inspirational stimuli for inferences across near and far fields to generate new design ideas by analogy and combination. J. Luo, S. Sarica, and K. L. Wood (Citation2019) presented a data-driven computer-aided rapid ideation process. InnoGPS provides guided retrievals of design information at the concept level in all known domains of technology based on the international patent classification according to their knowledge distance. Chen et al. (Citation2019) presented an integrated approach for enhancing design ideation by applying artificial intelligence and data mining techniques. The approach uses a semantic ideation network and a visual concepts combination model to provide inspiration. Han et al. (Citation2017) (Han, Hua, et al., Citation2020, Citation2018a, Citation2018b) presented several computational tools for supporting ideation based on semantic distance. Analogy Retriever (Han, Hua, et al., Citation2020) and Retriever (Han et al., Citation2018a) are computational tools for creative ideation based on analogical reasoning and ontology. The Analogy Retriever and the Retriever support reasoning for creative ideation through solving proportional analogy problems (A:B:C:X) by retrieving the unknown term X from a knowledge database to help designers construct an ontology. The Combinator (Han et al., Citation2018b) is a computer-based tool for supporting designers to produce creative ideas in idea generation. Combinator generates combinational prompts in text and image forms by combining unrelated ideas. Siddharth and Chakrabarti (Citation2018) developed a web-based, analogical design tool called Idea-Inspire 4.0. Idea-Inspire 4.0 is a searchable knowledge base of several biological systems and supports understanding and application of biological concepts in engineering design problems through analogical design.
许多基于语义距离（概念相似性）的计算方法和工具已经被开发出来，以支持构思。左等人（2022 年）提出了一个基于维基百科的语义网络，名为 WikiLink，这是一个用于设计创新的设计构思工具。为了为设计创新提供知识探索，WikiLink 采用了融合统计和语义关系的权重，捕捉了概念之间的隐含联系。康等人（2021 年）提出了一个支持工具 MetaMap，通过多维度基于示例的探索来激发视觉隐喻构思。为了促进构思过程，MetaMap 在语义、颜色和形状维度提供了基于示例的探索，基于关键词关联和颜色过滤提供了样本图像。罗、萨里卡和伍德（2021 年）提出了一个基于知识的专家系统，同时从概念、文档和领域层面提供了从总专利数据库中获取的设计刺激的计算引导探索和开发。 知识距离指导基于网络的探索和检索，以跨越近距离和远距离领域的启发性刺激进行类比和组合，生成新的设计理念。J. Luo，S. Sarica 和 K. L. Wood（2019）提出了一种数据驱动的计算机辅助快速构思过程。InnoGPS 根据知识距离，提供在所有已知技术领域中基于国际专利分类的概念级别设计信息的引导检索。Chen 等人（2019）提出了一种通过应用人工智能和数据挖掘技术来增强设计构思的综合方法。该方法利用语义构思网络和视觉概念组合模型提供灵感。Han 等人（2017）（Han，Hua 等人，2020，2018a，2018b）提出了几种基于语义距离支持构思的计算工具。类比检索器（Han，Hua 等人，2020）和检索器（Han 等人，2018a）是基于类比推理和本体论的创造性构思的计算工具。 类比检索器和检索器通过解决比例类比问题（A:B:C:X），从知识数据库中检索未知项 X，帮助设计师构建本体论，从而支持创意构思的推理。组合器（Han 等，2018b）是一种基于计算机的工具，用于支持设计师在创意生成中产生创意。组合器通过结合不相关的想法以文本和图像形式生成组合提示。Siddharth 和 Chakrabarti（2018）开发了一个基于网络的类比设计工具，名为 Idea-Inspire 4.0。Idea-Inspire 4.0 是一个可搜索的知识库，包含多个生物系统，并通过类比设计支持在工程设计问题中理解和应用生物概念。

Many studies have investigated the impact of analogical distance on the performance of ideation (Chan, Siangliulue, et al., Citation2017; Srinivasan et al., Citation2018). However, the effect of analogical distance on design outcomes is controversial (Jia et al., Citation2020). Some researchers argue that far-field analogies are more beneficial and others claim that near-field analogies are beneficial for ideation effectiveness (Jia et al., Citation2020). Han et al. Citation2017 (Han, Hua, et al., Citation2020; Han, Shi, et al., Citation2018) studied computational conceptual distances in combinational creativity. Studies indicate that far-related ideas are used more often than closely related ideas to produce creative combinational designs and far-related ideas could lead to more creative outcomes than closely related ideas. Chiu and Shu (Citation2012) showed that far-field analogies increase the novelty and quality of ideation. Kennedy et al. (Citation2018) showed that far-field analogies are beneficial on ideation effectiveness for novelty, relevance, and effectiveness of solutions. Taura and Nagai (Nagai & Taura, Citation2006; Taura et al., Citation2005; Taura & Nagai, Citation2013) studied concept generation based on analogical reasoning. Studies indicate that far-related base concepts could lead to higher originality outcomes in conceptual synthesis than closely related base concepts, and the synthesis of concepts in relationships between two base concepts associated with many other concepts leads to greater creativity. On the other hand, Srinivasan et al. (Citation2018) showed that near-field analogies are used more frequently, and they found that far-field analogies increase novelty but decrease quality. Chan, Siangliulue, et al. (Citation2017) examined competing theoretical recommendations, Associationist theory, and SIAM (Search for Ideas in Associative Memory) theory, for how inspirational delivery systems on collaborative ideation platforms should account for semantic distance of inspirational stimuli. The associationist theory of creativity (Gupta et al., Citation2012; Koestler, Citation1964; Mednick, Citation1962; Rothenberg, Citation1979) suggests that exposing ideators to ideas that are semantically far from their own maximizes novel combinations of ideas. On the other hand, the SIAM (Search for Ideas in Associative Memory) model of creative idea generation (Nijstad & Stroebe, Citation2006) cautions that systems should offer far ideas only when ideators reach an impasse and offer near ideas during productive ideation, which maximizes exploration within categories. The results show that far inspirations can be harmful for creativity if delivered during productive ideation and that collaborative inspiration systems could be improved by accounting for ideators’ cognitive states.
许多研究已经调查了类比距离对构思表现的影响（Chan，Siangliulue 等，2017 年；Srinivasan 等，2018 年）。然而，类比距离对设计结果的影响存在争议（Jia 等，2020 年）。一些研究人员认为远距类比更有益，而另一些人则认为近距类比有助于构思效果（Jia 等，2020 年）。Han 等人在 2017 年（Han，Hua 等，2020 年；Han，Shi 等，2018 年）研究了组合创造力中的计算概念距离。研究表明，远距离的想法比密切相关的想法更常用于产生创造性的组合设计，远距离的想法可能比密切相关的想法产生更具创意的结果。Chiu 和 Shu（2012 年）表明，远距类比增加了构思的新颖性和质量。Kennedy 等人（2018 年）表明，远距类比对构思的新颖性、相关性和解决方案的有效性有益。Taura 和 Nagai（Nagai 和 Taura，2006 年；Taura 等，2005 年；Taura 和 Nagai，2013 年）研究了基于类比推理的概念生成。 研究表明，远距离的基本概念可能会导致概念综合中更高的独创性结果，而不是与基本概念密切相关的情况，而且在两个基本概念之间的关系中综合概念与许多其他概念相关联会导致更大的创造力。另一方面，Srinivasan 等人（2018 年）表明，近场类比更常用，他们发现远场类比增加了新颖性但降低了质量。Chan，Siangliulue 等人（2017 年）检验了竞争性的理论建议，即联想主义理论和 SIAM（在联想记忆中搜索想法）理论，用于协作构思平台上的启发式交付系统应如何考虑启发性刺激的语义距离。创造力的联想主义理论（Gupta 等人，2012 年；Koestler，1964 年；Mednick，1962 年；Rothenberg，1979 年）表明，让构思者接触与他们自己的想法在语义上相距较远的想法，可以最大程度地实现想法的新颖组合。 另一方面，创意思维中的 SIAM（联想记忆中的思想搜索）模型（Nijstad & Stroebe，2006）警告称，系统应该在创意者陷入僵局时提供远距离的想法，并在生产性思维过程中提供近距离的想法，从而最大限度地探索各个类别。结果显示，如果在生产性思维过程中提供远距离的灵感，可能会对创造力产生不利影响，并且协作灵感系统可以通过考虑创意者的认知状态来改进。

In this study, we claim that AI inspiration based on conceptual similarity will significantly affect ideation. The use of conceptual similarity in this study builds on the SIAM theory, claiming semantically close ideas is more effective in ideation. AI-based conceptual similarity is measured in a latent space representation of word embeddings. The images presented to the designer have semantically close characteristics to the target design. To evaluate the effectiveness of AI-based conceptually similar inspirations, we have developed and applied metrics for design ideation.
在这项研究中，我们声称基于概念相似性的人工智能灵感将显著影响构思。本研究中对概念相似性的使用建立在 SIAM 理论基础上，声称在构思中语义上接近的想法更有效。基于人工智能的概念相似性是通过词嵌入的潜在空间表示来衡量的。呈现给设计师的图像具有与目标设计语义上接近的特征。为了评估基于人工智能的概念相似灵感的有效性，我们开发并应用了设计构思的度量标准。

2.3. Cognitive approaches to measuring ideation
2.3. 衡量构思的认知方法

Measuring ideation is a key to validating the claim that the AI inspiration based on conceptual similarity will influence design ideation in human-AI collaboration. There are several metrics for evaluating the performance of idea generation techniques such as fluency and novelty. The measures of ideation effectiveness can be based on the idea generation process or the originality of the final product (Borgianni et al., Citation2013).
衡量构思是验证基于概念相似性的人工智能灵感是否会影响人工智能协作中的设计构思的关键。有几种评估构思技术表现的指标，如流畅度和新颖性。构思有效性的衡量可以基于构思过程或最终产品的独创性（Borgianni 等，2013 年）。

Research on product-based approaches has resulted in several metrics: Sarkar and Chakrabarti (Citation2011) considered novelty and usefulness as measures of creativity. Shah et al. (Citation2003) introduced four metrics for measuring ideation effectiveness, used for evaluating idea generation in design: novelty, variety, quality, and quantity of designs. Nelson et al. (Citation2009) proposed a simple metric that combines novelty and variety to measure the amount and quality of design space exploration. Maher et al. (Citation2013) and Grace et al. (Citation2015) employed novelty, value, and surprise to evaluate design creativity. Taylor et al. (Citation1958) considered quantity and subjective assessments of the quality of the ideas as measures of ideation effectiveness. In this study, we adapt existing quantitative metrics for ideation (i.e., novelty, variety, and quantity) to evaluate the effect of a co-creative design system on design ideation.
基于产品的方法研究已经产生了几个度量标准：Sarkar 和 Chakrabarti（2011）认为新颖性和有用性是创造力的衡量标准。Shah 等人（2003）引入了四个用于衡量构思效果的度量标准，用于评估设计中的构思生成：设计的新颖性、多样性、质量和数量。Nelson 等人（2009）提出了一个简单的度量标准，结合了新颖性和多样性，以衡量设计空间探索的数量和质量。Maher 等人（2013）和 Grace 等人（2015）采用新颖性、价值和惊喜来评估设计创造力。Taylor 等人（1958）认为构思效果的衡量标准是构思的数量和对想法质量的主观评估。在本研究中，我们采用现有的定量构思度量标准（即新颖性、多样性和数量）来评估协同设计系统对设计构思的影响。

Cognitive-based approaches evaluate ideation processes based on cognitive processes inherent to creative thought. The Function-Behavior-Structure (FBS) ontology (J. S. Gero, Citation1990; Gero & Kannengiesser, Citation2004) is a design ontology that describes designed things, or artifacts, irrespective of the specific discipline of designing. The function (F) of a designed object is defined as its teleology; the behavior (B) of that object is either derived (Bs) or expected (Be) from the structure, where structure (S) represents the components of an object and their compositional relationships. These ontological classes are augmented by requirements (R) that come from outside the designer and description (D) that is the document of any aspect of designing. In this ontological view, the goal of designing is to transform a set of requirements and functions into a set of design descriptions. The transformation of one design issue into another is defined as a design process (J. S. Gero, Citation2010). We define an idea as a cognitive issue that the designer considers during the design process and adopt the Function-Behavior-Structure (FBS) ontology (J. S. Gero, Citation1990; Gero & Kannengiesser, Citation2004) as a basis for segmenting and coding each idea in the verbal protocol produced, while the participant is engaged in a design task.
基于认知的方法评估构思过程，基于创造性思维固有的认知过程。功能-行为-结构（FBS）本体论（J. S. Gero，1990；Gero & Kannengiesser，2004）是描述设计物品或工件的设计本体论，不考虑具体的设计学科。设计物品的功能（F）被定义为其目的论；该物品的行为（B）要么源自结构（Bs），要么期望自结构（Be），其中结构（S）代表物品的组成部分及其组成关系。这些本体类别由来自设计师外部的要求（R）和描述（D）增强，描述是设计的任何方面的文档。在这种本体论观点中，设计的目标是将一组要求和功能转化为一组设计描述。将一个设计问题转化为另一个设计问题被定义为设计过程（J. S. Gero，2010）。我们将一个想法定义为设计师在设计过程中考虑的认知问题，并采用功能-行为-结构（FBS）本体论（J. S.）。 Gero, 1990; Gero & Kannengiesser, 2004) 作为在参与设计任务时对口头协议中的每个想法进行分割和编码的基础。

3.1. System overview 3.1. 系统概述

The goal of CIP is to facilitate design ideation through AI-based inspiration by presenting conceptually similar design images to the user. From our exploratory study (J. Kim, M. L. Maher, & S. Siddiqui, Citation2021a, Citation2021b), we learned that the quality of the design images in the dataset is important in AI-based creativity. We also learned that inspirations based on conceptual similarity to the target design lead to more novel ideation than inspirations based on visual similarity to sketches drawn by a designer. Based on these observations, we update the CIP system to show only conceptually similar inspirations, high fidelity images of creative designs, and measure conceptual similarity based on multiple features of the description of the design task.
CIP 的目标是通过基于人工智能的灵感促进设计构思，向用户展示概念上类似的设计图像。根据我们的探索性研究（J. Kim, M. L. Maher, & S. Siddiqui, 2021a, 2021b），我们了解到数据集中设计图像的质量对基于人工智能的创造力至关重要。我们还发现，基于与目标设计的概念相似性的灵感比基于设计师绘制的草图的视觉相似性的灵感更有可能产生新颖的构思。基于这些观察，我们更新了 CIP 系统，仅展示概念上相似的灵感、创意设计的高保真图像，并根据设计任务描述的多个特征来衡量概念相似性。

The two main spaces of the user interface of the CIP are shown in
CIP 用户界面的两个主要空间如下所示:Figure 1 图 1: the design space (pink area) and the inspiring image space (purple area). The design space consists of a description of the design task, undo button, clear button, and the user’s canvas. The design task statement includes the name of the object to be designed as well as a context to further specify the objects’ use and environment. The user can sketch on the user’s canvas and edit the sketch using the undo and clear button. The inspiring image space includes an inspiring object name, ‘inspire me’ button, and the AI inspiration canvas. When the user clicks the ‘inspire me’ button, the AI model places an inspiring image in the AI inspiration canvas. Ideation using CIP is a cyclical process in which the user sketches, asks for inspiration, and the AI model presents an inspiring image. The title bar (gray area) of the user interface includes a hamburger menu, the name of the system, and an introductory statement about the CIP system. The hamburger menu allows the selection from one of the two design tasks, sink or bed, which allows the experiment facilitator to select one of the design tasks. The sink task results in a random selection of an inspiring image and the bed task results in a selection of an inspiring image based on conceptual similarity.
设计空间（粉色区域）和启发图像空间（紫色区域）。设计空间包括设计任务描述、撤销按钮、清除按钮和用户画布。设计任务说明包括待设计物体的名称以及进一步指定物体用途和环境的背景。用户可以在用户画布上草绘并使用撤销和清除按钮编辑草绘。启发图像空间包括启发物体名称、“启发我”按钮和 AI 启发画布。当用户点击“启发我”按钮时，AI 模型会在 AI 启发画布中放置一幅启发图像。使用 CIP 进行构思是一个循环过程，用户草绘，请求灵感，AI 模型呈现一幅启发图像。用户界面的标题栏（灰色区域）包括汉堡菜单、系统名称和关于 CIP 系统的介绍性说明。汉堡菜单允许从两个设计任务中选择一个，即水槽或床，这使实验引导者可以选择其中一个设计任务。 水槽任务会导致随机选择一幅鼓舞人心的图像，而床上任务会根据概念相似性选择一幅鼓舞人心的图像。

Figure 1. User interface of the Collaborative Ideation Partner (CIP).
图 1. 协作创意伙伴（CIP）的用户界面。

Display full size

Figure 2 图 2 shows a session with the CIP system in which the participant requests six inspiring designs and how their design sketch is modified in response to the inspiration. From the first inspiring image, the participant mimicked the table and legs from the inspiring image. The participant added a drawer as a new function, in response to the interior door in the second inspiring image. In this case, the compartments of the interior door were transferred to the drawer of the sink. From the third inspiring image, the participant came up with a piping system. The participant justified this change: ‘I thought about maybe like reusable water at first, but again I use anything you want to have clean water for your sink and brush your teeth and wash your hands or not. So I just added the piping system to the sink maybe the water could be used elsewhere but it is a piping system in a drain.’ The participant did not change anything based on the fourth or fifth image. From the sixth image which is a bubble tent, the participant noticed the reflections on the bubble tent and added a mirror. In summary, the interaction between the participant and the AI inspiration during ideation shows the participant mimicking a function or a shape from the inspiring images (first, second, and third inspiring image) and introducing new functions by reinterpreting the concept of inspiring objects (sixth inspiring image).
展示了与 CIP 系统的一个会话，参与者请求六个启发性设计，并展示了他们的设计草图如何根据启发进行修改。从第一个启发图像开始，参与者模仿了启发图像中的桌子和桌腿。参与者增加了一个抽屉作为新功能，以响应第二个启发图像中的内部门。在这种情况下，内部门的隔间被转移到了水槽的抽屉中。从第三个启发图像开始，参与者想出了一个管道系统。参与者对这一变化进行了解释：“一开始我考虑可能是可重复使用的水，但再次，我使用任何你想要的东西来获得清洁水，用于水槽刷牙和洗手或其他用途。所以我只是在水槽上添加了管道系统，也许水可以在其他地方使用，但它是一个排水系统。”参与者没有根据第四或第五个图像做出任何改变。从第六个图像，即一个气泡帐篷，参与者注意到了气泡帐篷上的反射，并添加了一个镜子。 总的来说，在构思过程中，参与者与人工智能灵感之间的互动表明参与者在模仿启发图像（第一、第二和第三个启发图像）中的功能或形状，并通过重新解释启发物体的概念（第六个启发图像）来引入新功能。

Figure 2. Example of ideation process using the Collaborative Ideation Partner (CIP).
图 2. 使用协作创意伙伴（CIP）进行构思过程的示例。

Display full size

3.2. Dataset of inspiring images
3.2. 激励图片数据集

For the source of inspiring designs, we collected a dataset of high-fidelity images of creative designs. To create the new dataset, we identified 20 object categories from 345 category sketches in QuickDraw! dataset (Jongejan et al., Citation2016) based on their conceptual similarity to the object in the design task (sink and bed). We then searched for images of five creative designs online for each object category using keywords ‘creative,’ ‘novel,’ ‘unusual,’ and ‘design’ (e.g., creative sink and unusual bed). The dataset thus contains 20 categories of objects with a total of 100 labeled images. Each image has three fields: id, object name, and design features. Id is the unique identifier that is assigned to each image. The object name is the name of the object that is represented in the image (e.g., electric massage bed, robotic advisor, and smart sofa). Design features are keywords that represent the design features and unique functionalities of the design (e.g., multi-functional, entertainment, massage, combinational, digital, and tv). These features were assigned by the research team.
为了激发设计灵感的来源，我们收集了一组高保真度的创意设计图像数据集。为了创建新的数据集，我们从 QuickDraw！数据集（Jongejan 等，2016）的 345 个类别草图中，基于它们与设计任务中物体的概念相似性，确定了 20 个物体类别（如水槽和床）。然后，我们使用关键词“创意”、“新颖”、“不寻常”和“设计”（例如，创意水槽和不寻常床）在线搜索每个物体类别的五个创意设计图像。因此，该数据集包含 20 个物体类别，共 100 张带标签的图像。每张图像有三个字段：id、物体名称和设计特征。Id 是分配给每个图像的唯一标识符。物体名称是图像中所代表的物体的名称（例如，电动按摩床、机器人顾问和智能沙发）。设计特征是代表设计特征和独特功能的关键词（例如，多功能、娱乐、按摩、组合、数字和电视）。这些特征由研究团队分配。

3.3. AI model for conceptual similarity
3.3. 概念相似性的人工智能模型

The AI model for conceptual similarity uses a deep-learning word embedding model to compute the degree of similarity (Karimi, Maher, et al., Citation2019) between a set of words in the design task statement and a set of words for each image in the image dataset. We generate a pairwise similarity score for each word in set 1 (words in the description of the design task) and each word in set 2 (words in the design feature list for each image). A Wikipedia pre-trained word2vec model is used to generate a vector representation for each of the words in both sets. We calculate the cosine similarity score for each pair of words for each image in the dataset. The similarity score for each image is calculated as the average of the pairwise cosine similarity scores. The larger number indicates that the two sets are more likely to appear in the same context, whereas a smaller number indicates that the two are less associated with each other. For example, a design task includes four words (i.e., bed, senior, living, and facility) and an image includes four words of design features (e.g., comfort, massage, combinational, and chair). For measuring the conceptual similarity between the design task and the image, we calculate the cosine similarity score for 16 pairs of words (4 words × 4 words) then calculate the average of these 16 scores. We construct the conceptual similarity ranking for each image based on its similarity score. When the participant requests inspiration, the system uses the ranking in order from the most conceptually similar to the least conceptually similar to select the next image.
概念相似性的人工智能模型使用深度学习词嵌入模型来计算设计任务陈述中一组词与图像数据集中每个图像的一组词之间的相似度（Karimi, Maher 等，2019）。我们为集合 1 中的每个词（设计任务描述中的词）和集合 2 中的每个词（每个图像的设计特征列表中的词）生成一对一相似度分数。我们使用维基百科预训练的 word2vec 模型为两个集合中的每个词生成向量表示。我们计算数据集中每个图像的每对词的余弦相似度分数。每个图像的相似度分数被计算为一对一余弦相似度分数的平均值。较大的数字表示两个集合更可能出现在相同的上下文中，而较小的数字表示两者之间的关联较小。例如，一个设计任务包括四个词（即床、高级、生活和设施），一个图像包括四个设计特征词（例如舒适、按摩、组合和椅子）。 用于衡量设计任务与图像之间概念相似性的方法，我们计算了 16 对词语（4 个词×4 个词）的余弦相似度分数，然后计算这 16 个分数的平均值。我们根据相似度分数构建每个图像的概念相似性排名。当参与者请求灵感时，系统将按照从最相似概念到最不相似概念的顺序使用排名来选择下一个图像。

4.1. Study design 4.1 研究设计

The experiment is a within-subject design that compares participants’ ideation, while engaged in a design task with different ideation stimuli: a control condition with random inspirations (condition A), a treatment condition with conceptually similar inspirations (condition B).
实验采用被试内设计，比较参与者在进行设计任务时的构思，使用不同构思刺激：一个控制条件是随机灵感（条件 A），一个处理条件是概念上相似的灵感（条件 B）。

• Condition A (control condition): randomly selected inspiration (sink)
  条件 A（对照条件）：随机选择的灵感（汇）

• Condition B (treatment condition): conceptually similar inspiration (bed)
  条件 B（治疗条件）：概念上类似的灵感（床）

During the study, for each participant and for each condition, we collected video protocol data during the design session and a retrospective protocol after the design session. The protocol including the informed consent document has been reviewed and approved by our IRB. We obtained informed consent from all participants. We recruited 55 university students (N = 55) for the participants: each participant engaged in both conditions.
在研究过程中，针对每位参与者和每种条件，我们在设计会话期间收集了视频协议数据，并在设计会话结束后进行了回顾性协议。协议包括知情同意文件已经经过我们的 IRB 审查和批准。我们从所有参与者那里获得了知情同意。我们招募了 55 名大学生（N = 55）作为参与者：每位参与者都参与了两种条件。

The task is an open-ended design task in which participants are asked to design an object in a given context through sketching. To reduce the learning effect, different objects for the design task were used for each condition: a sink for an accessible bathroom in the control condition and a bed for a senior living facility in the treatment condition. The design tasks for each condition were selected to have similar complexity and familiarity. The participants used a laptop and interacted with the CIP interface using a mouse to draw a sketch while performing the design task.
该任务是一个开放式设计任务，参与者被要求通过草图在特定背景下设计一个物体。为了减少学习效应，设计任务的不同条件使用了不同的物体：对照条件下使用了一个无障碍浴室的水槽，治疗条件下使用了一张老年居所的床。每个条件的设计任务被选为具有相似的复杂性和熟悉度。参与者使用笔记本电脑，并通过鼠标与 CIP 界面进行交互，在执行设计任务时绘制草图。

The procedure consists of a training session, two design task sessions, and two retrospective protocol sessions. In the training session, the participants are given an introduction to the features of the CIP interface and how to request inspiration from the AI model. After the training session, the participants perform the two design tasks. The study used a counterbalanced order for the two design tasks. The participants have no time limits to complete the design task and were instructed to perform the design task until they were satisfied with their design. The participants are free to click the ‘inspire me’ button as many times as they would like to get inspiration from the system. However, the participants were told to request at least three inspirational sketches during a single design task. The facilitator is present during the design task but does not interfere in the design process. Once a participant finishes the two design task sessions, the participant is asked to explain what they were thinking and how the inspiring images influenced their design while watching a recording of their design session.
该程序包括一次培训会议、两次设计任务会议和两次回顾性协议会议。在培训会议中，参与者将介绍 CIP 界面的特点以及如何向 AI 模型请求灵感。培训会议后，参与者执行两项设计任务。研究对这两项设计任务采用了平衡顺序。参与者在完成设计任务时没有时间限制，并被要求一直进行设计任务，直到他们对设计感到满意。参与者可以自由点击“启发我”按钮，以获取系统的灵感。然而，参与者被告知在单个设计任务期间至少要请求三次灵感草图。在设计任务期间，促进者在场但不干预设计过程。一旦参与者完成两次设计任务会议，就会要求参与者解释他们的思考过程以及启发性图像如何影响他们的设计，同时观看他们设计过程的录像。

4.2. Data collected 4.2. 数据收集

Two types of data were collected for analyzing the experiment’s results: a set of sketches that participants produced during the design tasks and verbalizing the ideation process during the retrospective protocol. We recorded the entire design task sessions and retrospective sessions for each participant. The sketch data shows the progress of the design ideation and the final design visually for each design task session. From the recordings of retrospective sessions, we collected verbal protocol data of the thoughts of the participants.
为分析实验结果，收集了两类数据：参与者在设计任务过程中制作的草图集以及在回顾性协议期间表达构思过程的语言。我们记录了每位参与者的整个设计任务会话和回顾性会话。草图数据展示了每个设计任务会话的设计构思进展以及最终设计的视觉呈现。从回顾性会话的录音中，我们收集了参与者的思考的语言协议数据。

4.3. Data segmentation and coding
4.3. 数据分割和编码

We transcribed the verbal data as a basis for analysis. The transcripts were segmented based on the presentation of inspiring images: An inspiring image segment starts when the participant requests an inspiring image and ends when the participant requests the next inspiring image. In this study, we define an idea as a cognitive issue using the FBS ontology (J. S. Gero, Citation1990; Gero & Kannengiesser, Citation2004). We further segmented the inspiring segments until each segment has a single code in the FBS ontology (J. S. Gero, Citation1990; Gero & Kannengiesser, Citation2004). An inspiring segment thus includes multiple-idea segments. After segmenting the verbal data, we conducted stemming, the process of reducing inflected words to their word stem. This stemming process allows us to identify unique ideas and repeat ideas in a design session.
我们将口头数据转录为分析的基础。转录根据激发性图片的呈现进行分割：一个激发性图片片段从参与者请求激发性图片开始，直到参与者请求下一个激发性图片结束。在这项研究中，我们使用 FBS 本体论定义思想为认知问题（J. S. Gero, 1990; Gero & Kannengiesser, 2004）。我们进一步将激发性片段分割，直到每个片段在 FBS 本体论中具有单一代码（J. S. Gero, 1990; Gero & Kannengiesser, 2004）。因此，一个激发性片段包括多个思想片段。在分割口头数据后，我们进行了词干处理，即将屈折词减少到其词干。这个词干处理过程使我们能够在设计会话中识别独特的思想和重复的思想。

Table 2 表 2 presents an example of data segmentation and coding for an inspiring image. The example includes an inspiring image segment (i.e., second column) for the ‘functional sleeping bag’ image (i.e., first column) and the inspiring image segment is segmented into 12 idea segments (i.e., third column). Each idea segment is then summarized as a single word that describes the idea (fourth column).
展示了一个关于激发性图像的数据分割和编码示例。该示例包括一个激发性图像段（即第二列）用于“功能睡袋”图像（即第一列），并且激发性图像段被分成 12 个想法段（即第三列）。然后，将每个想法段总结为描述该想法的单个词（第四列）。

Two researchers coded 110 sessions of retrospective protocol (i.e., 55 sessions of condition A and 55 sessions of condition B) individually based on FBS ontology, then came to consensus for the different coding results. The coding instruction given to the coders included how to segment inspiring sketch segments and idea segments. The two coders segmented and coded a design session together to make an initial agreement for segmentation and coding before coding individually, then coded all design sessions individually. Once each coder completed coding all data individually, the two coders discussed each of the different coding results and came to consensus.
两位研究人员根据 FBS 本体论对 110 个回顾性协议会话进行编码（即 55 个 A 条件会话和 55 个 B 条件会话），然后就不同的编码结果达成共识。给予编码人员的编码说明包括如何分割启发式草图片段和想法片段。两位编码人员一起分割和编码一个设计会话，以在单独编码之前就分割和编码达成初步协议，然后单独对所有设计会话进行编码。一旦每位编码人员单独完成所有数据的编码，两位编码人员讨论每个不同的编码结果并达成共识。

5.1. Temporal analysis 5.1. 时间分析

We hypothesize that the quantity in novelty, variety, and quantity of ideas over time decreases more slowly in an ideation with AI-based inspirations than the temporal pattern of ideation with random inspirations. Prior research shows a time effect on idea generation when the rate of idea generation decreases over time (Howard-Jones & Murray, Citation2003; Liikkanen et al., Citation2009; Perttula & Liikkanen, Citation2006; Snyder et al., Citation2004; Tsenn et al., Citation2014). We posit that the effect of AI-based inspirations based on conceptual similarity affects the time effect on design ideation.
我们假设，在基于人工智能灵感的构思中，随着时间的推移，新颖性、多样性和想法的数量减少得比基于随机灵感的构思的时间模式更缓慢。先前的研究表明，在想法生成过程中存在时间效应，即想法生成的速率随时间减少（Howard-Jones & Murray, 2003; Liikkanen 等，2009; Perttula & Liikkanen, 2006; Snyder 等，2004; Tsenn 等，2014）。我们认为，基于概念相似性的人工智能灵感对设计构思的时间效应产生影响。

Figure 3 图 3 shows the number of novel ideas for the first 30 segments in the control condition and the treatment condition. The results of temporal patterns show that the quantity of novel ideas decreases over time in both the control condition and the treatment condition as we expected. However, there are two distinct patterns between the results of the control condition and the treatment condition. The changes of idea number between the ‘before’ segment and the first segment show a contradictory result between the control condition and the treatment condition. In the control condition, the number of ideas decreased in the first segment. However, in the treatment condition, the number of ideas increased in the first segment. In the ‘before’ segment, participants produced many ideas without any inspiring images since they generate ideas to meet the basic requirements of the target design as a basic concept of their design. This result indicates that the conceptually similar inspiration leads to more novel ideas. The second pattern we found is the distribution of the ideas. Although the results of both the control condition and the treatment condition show a decreasing pattern, the distribution until the 20th segment shows a significant difference between the control condition and the treatment condition.
显示了对照条件和治疗条件中前 30 个片段的新颖想法数量。时间模式的结果显示，新颖想法的数量随时间在对照条件和治疗条件中都如我们所预期的那样逐渐减少。然而，在对照条件和治疗条件的结果之间存在两种明显的模式。在‘之前’片段和第一个片段之间的想法数量变化在对照条件和治疗条件之间呈现出矛盾的结果。在对照条件中，第一个片段中的想法数量减少。然而，在治疗条件中，第一个片段中的想法数量增加。在‘之前’片段中，参与者产生了许多没有任何启发图像的想法，因为他们生成想法以满足目标设计的基本要求作为设计的基本概念。这一结果表明，在概念上相似的灵感会带来更多新颖的想法。我们发现的第二种模式是想法的分布。 尽管对照组和治疗组的结果都显示出下降的趋势，但在第 20 个片段之前的分布显示出对照组和治疗组之间的显著差异。

Figure 3. The number of novel ideas in each inspiring image segment.
图 3. 每个启发性图像部分中新颖想法的数量。

Display full size

To visualize the differences between the temporal patterns in the two conditions, we generated trend lines with a scatter plot and a comparison shows a difference between the control condition and the treatment condition (
为了可视化两种条件下时间模式之间的差异，我们利用散点图生成了趋势线，并比较显示了对照组和治疗组之间的差异Figure 4 图 4). A paired t-test was conducted to determine the significance of the difference between the control condition and the treatment condition in temporal changes of novelty. The quantity of novel ideas over time decreased more slowly in the treatment condition (M = 18.17, SD = 21.52) than in the control condition (M = 14.60, SD = 19.68), t(54) = 1.69, two tail M = 2.46E–11.
进行了成对样本 t 检验，以确定控制条件和处理条件在新颖性时间变化方面的差异的显著性。随着时间推移，处理条件中新颖想法的数量下降速度比控制条件慢（M = 18.17，SD = 21.52），t(54) = 1.69，双尾 M = 2.46E–11。

Figure 4. The trendlines of novelty.
图 4. 新颖性的趋势线。

Display full size

Figure 5 图 5 shows the number of new ideas for the first 30 segments in the two conditions. The results of temporal patterns are similar to the results of novelty in terms of decreasing patterns with the time effect and the increasing pattern of the treatment condition in the first segment. However, the increasing pattern of the treatment condition in the first segment is relatively less than the results of novelty and the result of the treatment condition after the first segment shows a similar pattern to the control condition largely decreasing in the second segment.
显示了两种条件下前 30 个片段中新想法的数量。在时间模式方面的结果与新颖性方面的结果相似，都表现出随着时间效应的减少模式和治疗条件在第一个片段中的增加模式。然而，治疗条件在第一个片段中的增加模式相对较少于新颖性的结果，第一个片段后治疗条件的结果显示出与控制条件类似的模式，在第二个片段中大幅减少。

Figure 5. The number of new ideas in each inspiring image segment.
图 5. 每个启发性图像部分中新想法的数量。

Display full size

Figure 6 图 6 shows the results as trend lines with a scatter plot for the temporal patterns of variety. Visually, we see a difference from the first segment to the 20th segment, but not after the 20th segment. A paired t-test was conducted and shows a significant difference between the control condition and the treatment condition in temporal changes of variety. The quantity in variety of ideas over time decreased more slowly in the treatment condition (M = 123.78, SD = 185.79) than in the control condition (M = 92.79, SD = 144.15), t(54) = 1.69, two tail M = 0.00039.
展示了种类时间模式的趋势线和散点图结果。从第一个部分到第 20 个部分，我们在视觉上看到了差异，但在第 20 个部分之后就没有了。进行了成对 t 检验，结果显示在种类时间变化方面，控制条件和处理条件之间存在显著差异。随着时间推移，处理条件下的想法种类数量下降速度比控制条件下的更慢（M = 123.78，SD = 185.79），控制条件下为（M = 92.79，SD = 144.15），t(54) = 1.69，双尾 M = 0.00039。

Figure 6. The trendlines of variety.
图 6. 品种的趋势线。

Display full size

Figure 7 图 7 shows the number of ideas for the first 30 segments in the two conditions. The results of temporal patterns are similar to the results of novelty and variety. In the treatment condition, the number of ideas was largely increased in the first segment like the result of novelty. The results after the first segment show a similar pattern to the results of variety with differences between the control condition and the treatment condition.
显示了两种条件下前 30 个片段的想法数量。时间模式的结果与新颖性和多样性的结果类似。在处理条件下，第一个片段的想法数量大幅增加，类似于新颖性的结果。第一个片段之后的结果显示出与多样性结果类似的模式，在控制条件和处理条件之间存在差异。

Figure 7. The number of ideas in each inspiring image segment.
图 7. 每个启发性图像部分中的想法数量。

Display full size

Figure 8 图 8 shows the results as trend lines with a scatter plot for the temporal patterns of quantity. The results show a similar pattern to the results of the variety, showing a difference from the first segment to the 20th segment. The result of a paired t-test that compares the trend lines showed a significant difference between the two conditions. The quantity in quantity of ideas over time decreased more slowly in the treatment condition (M = 226.19.78, SD = 307.37) than in the control condition (M = 175.24, SD = 248.33), t(54) = 1.69, two tail M = 0.000067.
显示结果为数量的时间模式趋势线和散点图。结果显示与品种结果类似的模式，显示从第一段到第 20 段的差异。比较趋势线的配对 t 检验结果显示两种条件之间存在显著差异。随时间推移，治疗条件下的思想数量减少速度比对照条件慢（M = 226.19.78，SD = 307.37），t(54) = 1.69，双尾 M = 0.000067。

Figure 8. The trendlines of quantity.
图 8. 数量的趋势线。

Display full size