在以“立德树人”为根本任务、培养德智体美劳全面发展的社会主义建设者和接班人的教育目标指引下，坚毅力是学生综合素质中必备的优秀品质之一，也是我国未来人才培养的核心目标之一，对其开展测评具有重要的现实意义。然而，既有的坚毅力测评工具主要为量表、方法主要为自我报告法，测评依托的数据以单一模态数据为主，不能得到较为客观、真实的测评结果。鲜有基于探究实践活动场景真实表现的测评工具与智能测评方法体系。多模态学习分析技术的兴起与不断成熟发展使得我们能够获取面向场景真实表现的多模态数据，支持坚毅力测评实现突破性创新；另外，随着表现性评价的推进，探究实践活动已然是坚毅力测评的重要场景，支持测评往纵深方向发展，有望解决学生坚毅力测评现存的现实问题。

因此，如何基于特定场景的学生多模态真实表现数据实现对坚毅力的客观化与智能化测评成为本研究需要解决与破解的核心问题，具体分为四个子研究问题：学生坚毅力测评的理论模型是什么？如何设计基于科学探究活动场景的学生坚毅力测评工具及数据指标？如何对学生坚毅力测评的多模态数据进行计算？基于测评模型的学生坚毅力表现如何？基于以上四个子研究问题，研究相对应地开展了四方面的内容研究。

首先，研究立足于现有相对宽泛的学生坚毅力测评理论框架的基本判断，借鉴扎根理论的基本思想与方法，采用NVivo 11软件对84篇有关坚毅力研究的国内外权威期刊文献与报告进行开放式编码、关联式编码以及核心式编码分析，批判性构建学生坚毅力测评初始理论模型。基于初始理论模型，研究采用德尔菲法，设计专家函询问卷，邀请15位专家开展了三轮论证，并对学生坚毅力测评各级指标进行赋权。最终，基于论证结果分析，研究构建了包含坚毅的行为性、坚毅的情感性、坚毅的认知性等3个一级指标，专注性、坚持性、积极情感、消极情感、目标意识、自我监控等6个二级指标的学生坚毅力测评理论模型。

其次，研究以理论模型为指导，融入表现性评价理念，从基于信息技术的新型测评方式视角出发，制定工具设计原则与设计思路，整合中国科学技术馆现有科学探究活动资源，初始设计由“学习视频+学习任务单”两部分组成的学生坚毅力测评工具。为保证工具的科学性与可靠性，研究采用自上而下的专家咨询方式对工具内容设计进行科学性评估，对表现性任务质量进行质量评价，形成第一轮修订工具。进一步，基于该工具，研究采用自下而上的方法采集368名试测学生在科学探究活动场景下的坚毅力多模态数据。研究运用IRT对采集的数据进行验证分析，结合学生、家长、教师与专家咨询，对工具进行第二轮修改，形成了由“目标信息+学习视频+学习任务单”三部分组成的学生坚毅力测评工具。基于以上，研究设计了17个表征学生坚毅力测评二级指标的初始多模态数据指标，同时设计专家函询问卷，邀请8位领域专家对多模态数据指标进行论证与修订。基于论证结果分析，研究最终保留16个与学生坚毅力测评二级指标相对应的多模态数据指标。

然后，研究按照“数据采集与筛选→特征构建与提取→模态融合与指标计算→模型生成与结果计算”的流程构建了学生坚毅力测评多模态数据表征模型，并进行了实践有效性的验证。第一，研究采集与筛选511名B大学附属学校四年级学生的坚毅力多模态数据，包括视频数据（头部姿态、行为动作序列、面部表情）、图像数据（自我反思报告）、文本数据（自我反思报告、目标信息题项答案）以及平台日志数据（系统行为日志）；第二，基于多模态数据指标体系，研究构建22个细化特征，并采用统计分析、TF-IDF、BERT、TextRank、Yolov7等算法、技术分别提取日志数据特征、文本数据特征以及视频数据特征；第三，研究采用基于熵权法的决策融合策略对理论模型每个二级指标的多模态数据进行融合，并使用加权求和的方式，计算每个二级指标的结果，包括专注性得分、坚持性得分、积极情感得分、消极情感得分、目标意识得分、自我监控得分；第四，以理论模型为依据，研究通过自下而上逐级加权求和的方式，生成可计算模型，计算坚毅的行为性得分、坚毅的情感性得分、坚毅的认知性得分，以及学生坚毅力指数。为验证可计算模型的有效性，研究将模型计算结果与他评所得511名学生坚毅力总分结果进行相关分析与回归分析。结果表明，可计算模型具有实践有效性。

最后，研究基于测评模型，开展了以B大学附属学校486名四年级学生群体为规模的学生坚毅力表现分析。一是采用描述性统计分析方法呈现学生坚毅力表现总体情况，二是采用方差分析方法比较不同人口学特征、不同类型学校以及高低分组学生坚毅力表现的差异，三是采用K-means聚类算法刻画学生坚毅力表现的类型特征。基于上述分析，研究得到如下结论，整体上，在科学探究活动场景下：（1）四年级学生坚毅力指数整体处于偏低水平，且各级指标发展水平差异大；（2）四年级学生坚毅力发展在性别、是否独生子女、不同类型学校上存在一定差异，但是差异均较小；（3）高低分组的四年级学生坚毅力表现存在不同程度的差异，且效应量强、差异明显；（4）四年级学生坚毅力表现存在五种类型的特征刻画，分别为：潜力型坚毅力者、全面型坚毅力者、行为型坚毅力者、认知型坚毅力者、行为与认知并重型坚毅力者，为未来学生坚毅力的发展与培养提供了一定的参考价值。此外，研究基于初步的结论，针对学生坚毅力的发展与培养，从学校、教师、学生以及教育研究者等不同层面提出了不同的发展建议。

通过对上述问题的探索，研究解决了当前学生坚毅力测评客观性不足、真实性不够等现实问题，并重点在测评工具、测评方法等两个方面实现了创新：在测评工具上，设计并开发面向科学探究实践活动场景真实表现的学生坚毅力测评工具；在测评方法上，提出基于多模态数据的相对客观化的学生坚毅力智能测评方法体系。研究所取得的成果不仅能够从理论层面丰富当前坚毅力的内涵、结构，还能够从实践层面提供测评工具设计方案指导与多模态数据支持的智能化测评方法体系，为学生未来的个性化坚毅力培养提供依据，为同类测评提供借鉴，助力破解教育评价改革的“卡脖子”问题，具有较好的科学价值与应用价值。

Under the guidance of the fundamental task of cultivating morality and nurturing talented individuals and the goal of fostering well-rounded socialist constructors and successors who excel in moral, intellectual, physical, aesthetic, and labor development, grit is one of the essential qualities in students’ comprehensive quality. It is also one of the core objectives in the cultivation of future talents in China. Therefore, the assessment of students’ grit holds important practical significance. However, existing grit assessment tools mainly consist of scales and methods primarily rely on self-reporting, with primarily based on single-modal data, which fails to yield sufficiently objective and authentic assessment results. There is a lack of assessment tools and intelligent assessment methodologies based on authentic performance in inquiry activities. The emergence and continuous maturation of multimodal learning analytics technology enable us to obtain multimodal data reflecting authentic performance in context, facilitating breakthrough innovations in grit assessment. Furthermore, with the advancement of performance assessment, inquiry activities have been important scenarios for grit assessment, supporting the deepening of assessment. The above is expected to address the current practical issues in grit assessment for students.

Therefore, how to conduct an objective and intelligent assessment of students’ grit based on real multimodal data in specific scenarios becomes the core issue that this study needs to address and resolve. Specifically, it is divided into four sub-research questions, that is, what is the theoretical model of students’ grit assessment? how to design the assessment tools and data indicators for students’ grit assessment in the context of scientific inquiry activities? how to compute multimodal data for students’ grit assessment? how do students’ grit perform based on the assessment model? Based on these four sub-research questions, corresponding content research has been conducted in four aspects.

Firstly, based on the fundamental assessment of the relatively broad theoretical framework for evaluating students’ grit and drawing on the basic principles and methods of grounded theory, the study utilized NVivo 11 software to conduct open, axial, and selective coding analyses of 84 authoritative journal articles and reports related to grit from both domestic and international sources. Through critical examination, an initial theoretical model for assessing students’ grit was developed. Based on this initial theoretical model, the study employed the Delphi method, designing expert consultation questionnaires and inviting 15 experts to participate in three rounds of deliberation to weigh the various levels of indicators for assessing students’ grit. Ultimately, through analysis of the deliberation results, the study constructed a theoretical model for evaluating students’ grit, consisting of three primary indicators: the behavior of grit, the emotion of grit, and the cognition of grit. These primary indicators were further delineated into six secondary indicators, including concentration, perseverance, positive emotion, negative emotion, goal awareness, and self-control.

Next, guided by the theoretical model and integrating the concept of performance assessment, the study adopted a perspective of innovative assessment methods based on information technology to formulate principles and approaches for tool design. Furthermore, the study integrated existing scientific inquiry resources from the China Science and Technology Museum to develop an initial students’ grit assessment tool consisting of learning videos plus learning task sheets. To ensure the scientificity and reliability of the tool, the study employed a top-down approach of expert consultation to scientifically evaluate the content and the quality of performance tasks, resulting in the first revision. Based on the first version tool, the study adopted a bottom-up approach to collect multimodal data of grit from 368 pilot-tested students in scientific inquiry activities. By utilizing the Item Response Theory (IRT) for validation analysis, and incorporating consultations with students, parents, teachers, and experts, a second revision tool was done, resulting in the formation of goal information, learning videos, and learning task sheets. Based on this, the study designed 17 initial multimodal data indicators representing secondary indicators of the assessment theoretical model of students’ grit. Simultaneously, expert consultation questionnaires were designed, and 8 domain experts were invited to deliberate and revise the multimodal data indicators. Through analysis of the deliberation results, the study ultimately retained 16 multimodal data indicators.

Then, following the process of data collection and screening, feature construction and extraction, modality fusion and indicator calculation, and model generation and result calculation, the study constructed a multimodal data representation model for assessing students’ grit and verified its practical effectiveness. First, the study collected and screened the multimodal data of grit from 511 fourth-grade students at B University Affiliated School. This data included video data (head posture, behavioral action sequences, facial expressions), image data (self-reflection reports), text data (self-reflection reports, answers to goal information questions), and platform log data (system behavior logs). Second, based on the multimodal data indicator system, the study developed 22 refined features and utilized statistical analysis, TF-IDF, BERT, TextRank, Yolov7 algorithms and technologies to extract features from log data, text data, and video data separately. Third, the study employed a decision fusion strategy based on the entropy weight method to fuse multimodal data for each secondary indicator of the theoretical model. Using weighted summation, the study calculated results for each secondary indicator, including concentration score, perseverance score, positive emotion score, negative emotion score, goal awareness score, and self-control score. Fourth, based on the theoretical model, the study generated a computable model through a bottom-up weighted summation approach to calculate the behavior of grit score, the emotion of grit score, the cognition of grit score, and students’ grit index. To verify the effectiveness of the computable model, the study conducted correlation and regression analyses between the model-calculated results and the externally assessed total grit scores of the 511 students. The results indicated that the computable model demonstrated practical effectiveness.

Finally, based on the assessment model, the study analyzed students’ grit performance using a cohort of 486 fourth-grade students from B University Affiliated School. Firstly, descriptive statistical analysis was employed to present the overall situation of students’ grit performance. Secondly, analysis of variance (ANOVA) was used to compare differences in grit performance among students with different demographic characteristics, school types, and high/low-scoring groups. Thirdly, the K-means clustering algorithm was applied to characterize the types of students’ grit performance. Based on the above analyses, the following conclusions were drawn regarding students’ grit performance in the context of scientific inquiry activities. (1) Overall, fourth-grade students exhibited relatively low levels of grit index, with significant differences in the development levels of various indicators. (2) Differences in grit development were observed among fourth-grade students based on gender, whether they were only children, and school types, but these differences were relatively small. (3) There were significant differences in grit performance between high/low-scoring groups of fourth-grade students, with strong effect sizes and noticeable differences. (4) Five types of grit performance were identified among fourth-grade students, namely: potential grit performers, comprehensive grit performers, grit performers with behavior, grit performers with cognition, and grit performers with balanced behavior and cognition, providing valuable insights for the future development and cultivation of students’ grit. Additionally, based on the preliminary conclusions, the study provided development recommendations for fostering students’ grit from the perspectives of schools, teachers, students, and educational researchers.

Through the exploration of the aforementioned issues, the study has addressed the practical problems of insufficient objectivity and authenticity for students’ grit assessments. The study has focused on innovation in two aspects, the assessment tools and the assessment methods. In terms of assessment tools, a students’ grit assessment tool has been designed and developed, which is oriented towards real performance in scientific inquiry practical activities. Regarding assessment methods, an intelligent assessment method system for students’ grit has been proposed, which is based on multimodal data and relatively objective. The achievements of the study not only enriched the connotation and structure of grit at the theoretical level, but also provided guidance for assessment tool design and intelligent assessment method systems based on multimodal data at the practical level. These findings serve as a basis for personalized grit cultivation in students’ future education and offer insight for similar assessments, aiding in overcoming the bottleneck issues in educational evaluation reform. The study holds significant scientific and practical value.