近年来，各种组织对于创造性人才有着巨大的需求，创造性人才测量的一个重点是创造性人格的测量。但是，创造性人格的测量在一段时间中依赖于外显测量工具，在组织招聘等场景中使用外显创造性人格测量容易受到求职者“装好”的影响导致结果失真，因此各组织对有一定抗作弊能力的创造性人格测量工具有较大的需求。本研究就试图通过使用图片故事练习测量内隐创造性人格的方式来满足这一需求。图片故事练习是一种内隐测量工具，它通过收集被试对图片刺激的反应（故事）来对被试的内隐特质进行判断。本研究结合前人文献，使用实证研究的方式来构造创造性人格图片故事练习及其评分系统。

本研究基于资料搜集、访谈等方法初步建立了大学生创造性人格图片故事练习的评分结构，然后根据访谈内容与专家审核，本研究选择了最适合的五张图片用以引发被试的反应，接下来，本研究进行了两轮编码以开发最终版的大学生创造性人格图片故事练习评分结构，剔除了没有让被试产生反应得两张图片。最终本测验展现出了尚可接受的信效度水平，总的来说，科隆巴赫α系数为0.48，评分者共同率为0.83，与内隐创造性人格量表的相关为0.31，但是与创造性行为量表不存在相关。

此外，Schultheiss（2020）建议将有监督的机器学习应用于图片故事练习之中以提高测验评分的效率，本研究也试图利用中文短文本机器学习分类技术开发创造性人格图片故事练习的自动评分系统。该系统能够针对每个句子是否有创造性人格特征进行二分分类，然后根据图片故事练习的计分规则求得被试在创造性人格图片故事练习上的总分。基于此前研究开发的大学生创造性人格图片故事练习，使用Python进行文本预处理（分词及处理停用词），之后采用one-hot编码将自然语言转换为机器语言并使用词频-逆文档频率（Term Frequency-Inverse Document Frequency, TF-IDF）模型对短文本特征进行加权。接着选择最常用的支持向量机、随机森林机器学习算法训练模型，最后得到了模型拟合结果。根据模型结果，随机森林模型的效果最优。总的来说，准确率为0.87，召回率较低只有0.37，Marco-F1为0.72。

总的来说，大学生创造性人格图片故事练习展现出了合格的信效度水平，就相容效度而言，大学生创造性人格图片故事练习与内隐创造性人格量表存在显著正相关，就校标效度而言，大学生创造性人格图片故事练习与创造性行为量表不存在相关。根据大学生创造性人格图片故事练习开发的自动评分系统能够准确识别阳性语料，但是该自动评分系统容易错误分类阴性语料。

In recent years, various organizations have a huge demand for creative talents. One of the key points of creative talent’s measurment is the measurement of creative personality. However, the measurement of creative personality depends on explicit measurement for a period of time. The use of explicit creative personality measurement in organizational recruitment is vulnerable to the influence of job seekers' “faking”, resulting in distorted results. Therefore, organizations have a great demand for creative personality measurement tools with anti cheating ability. This study attempts to meet this need by using picture story practice to measure implicit creative personality. Picture story exercise is an implicit measurement tool, which judges the implicit characteristics of subjects by collecting their responses to picture stimuli. Combined with previous literature, this study uses empirical research to construct implicit creative personality picture story practice and its scoring system.

In this reasearch, based on data collection, interviews and other methods, a scoring structure for the picture story exercise of College Students' implicit creative personality was initially established. Then, according to the interview content and expert review, this study selected the most suitable five pictures to trigger the reaction of the subjects. Next, this study conducted two rounds of coding to develop the final version of the scoring structure for the picture story exercise of College Students' implicit creative personality, Two pictures that did not make the subjects react were excluded. Finally, the test showed an acceptable level of reliability and validity. In general, The cronbach α coefficient was 0.48, the common rate of raters was 0.83, and the correlation with the implicit creative personality scale was 0.31, but there was no correlation with the creative behavior scale.

In addition, Schultheiss (2020) suggested that supervised machine learning be applied to picture story exercise to improve the efficiency of test scoring. In the supplementary study, this study tried to develop an automatic scoring system for implicit creative personality picture story practice by using Chinese short text machine learning classification technology. The system can classify whether each sentence has creative personality characteristics, and then obtain the total score of the subjects' implicit creative personality picture story practice according to the scoring rules of the picture story practice. The supplementary study is based on the picture story exercise of college students' implicit creative personality developed in previous reasearch. Python is used for text preprocessing and then one hot coding is used to convert natural language into machine language, and term frequency inverse document frequency (TF-IDF) model is used to weight the characteristics of short text. Then select the most commonly used support vector machine and random forest machine learning algorithm to train the model, and finally get the model fitting results. According to the model results, the effect of random forest model is the best. Overall, the accuracy rate is 0.87, the recall rate is low, only 0.37, and marco-f1 is 0.72.

In general, college students' implicit creative personality picture story exercise shows a qualified level of reliability and validity. In terms of compatibility validity, college students' implicit creative personality picture story practice has a significant positive correlation with the implicit creative personality scale. In terms of calibration validity, college students' implicit creative personality picture story exercise has no correlation with the creative behavior invetory. The automatic scoring system developed base on the implicit creative personality picture story exercise can accurately identify the positive sentences, but the automatic scoring system is easy to misclassify the negative sentences.