来华留学生跨文化适应性规则提取研究

【作者】 王冬燕；

【导师】 余嘉元；

【作者基本信息】 南京师范大学 ， 基础心理学， 2013， 博士

【摘要】 为了探讨来华留学生跨文化适应性影响因素与适应性水平之间的关系,通过文献综述、访谈法和问卷法等,获得了来华留学生在人口统计学因素、适应性影响因素和适应性水平三大部分情况。利用全模型结构方程、关联规则方法、分类回归树(CART)方法,从多个角度对来华留学生跨文化适应性进行了分析。研究结果充实了当前留学生适应性理论,并首次成功地将数据挖掘方法应用于留学生适应性的研究。方法论上,不仅在来华留学生研究领域是个突破,而且还为心理测量中的数据分析方法扩充了新的内容。主要研究结论有以下几点：(1)预测试后,通过探索性因素分析和验证性因素分析,确定了适应性影响因素和适应性水平正式问卷各个部分的结构。对来华留学生人口统计学因素进行方差分析,研究结果表明：留学生来自的国家文化、经济情况都对适应性产生影响,而且留学生的性别、来华时间以及来华之前对华了解的情况,都是影响其适应性的主要因素；留学原因对适应性影响也很大,主要表现在获得奖学金、学费低、喜欢汉语、想了解中国这几个方面。(2)采用结构方程建模的方式建立留学生适应影响因素与适应性水平的全模型结构,研究结果表明：模型整体拟合程度及解释能力基本能满足要求,内容效度与结构效度较高,较为全面系统地反映了来华留学生适应性影响因素与适应性水平之间的关系。(3)将关联规则方法引入来华留学生跨文化适应性的研究。采用Visual C++编程环境实现了关联规则Apriori算法,挖掘出了频繁2-项集规则。将最小支持度和最小置信度设定为0.1和0.6,然后根据不同的支持度和置信度,共进行17次实验,实验结果获得524条关联规则。(4)将频繁2-项集规则结果与传统相关分析方法进行比较。研究结果表明：频繁2-项集关联规则可以很好的表达相关分析结果,甚至可以推出相关分析所没有表现的相互关系,相关分析结果可以看作关联规则方法分析的子集,关联规则方法还可以在一定程度上表现出比相关更强的相互关系。由此可见,关联规则方法可有效应用于心理测量数据相互关系的分析。(5)利用WEKA软件挖掘频繁多项集规则,共进行6次实验,根据不同支持度和置信度获得不同数量的频繁多项集规则。研究发现：有些变量的频繁多项集规则的置信度,比频繁2-项集规则的置信度要高,在一定程度上加强了变量之间的相互关系,提供更多变量之间的相关信息。(6)将决策树方法用于来华留学生跨文化适应的研究,选用CART算法对留学生适应性进行分类。研究结果：建立了总适应性、社会文化适应、心理适应和校园适应四个CART树模型,共提取29条分类规则。(7)将CART分类规则与传统的二元Logistic回归分析进行比较,并通过分类正确率和ROC曲线验证两种分类模型的效能。结果发现,两种分类模型各有利弊,检验参数也十分相近。CART分类在挖掘主效应因素,并获得规则方面,与Logistic回归产生交叉因素,共同验证了服务模式、外向性、教师形象和学习条件是具有最佳预测效果的因子。因此可以说,决策树方法能够有效用于心理测量数据分类的研究,并且,CART分类获得的变量比Logistic回归获得的要多,这与决策树挖掘较为精细、能够避免自变量共线性的影响有很大关系。更多还原

【Abstract】 In order to explore the relationship between the foreign students’cross-cultural adaptation influencing factors and adaptive level in China, getting the information about foreign students’demographic factors, adaptive influencing factors and adaptive level through literature review, interviews and questionnaires. Using all model structural equation models, association rules, and classification and regression tree method, to analyze cross-cultural adaptation of foreign students from more than one angle. Findings enriched current international students’adaptability theory, and firstly successful applied the data mining method to the study of foreign students’ adaptability. Methodology, it’s not only a breakthrough in the field of foreign students, but also a new psychometric data analysis methods. The main conclusions are the following:1) After pre-tested, through exploratory factor analyzed and confirmatory factor analyzed to determine the various parts of the formal questionnaire including the influencing factors of adaptability and adaptive level structure. Variance analyzed of demographic factors on foreign students, the results showed as follows:Foreign students’countries cultural and economic circumstances impacted on adaptability, and the gender of the students, the time length in China and the situation of understanding China before coming to China are the main factors to affect adaptability. Studying factors also is a great impact on adaptability, main performance in getting scholarships, low tuition, liking Chinese, and want to learn Chinese.2) Used structural equation modeling to establish the model structure between influencing factors of adaptability and adaptive level of foreign students, the results showed as follows:The whole model fitting degree and explanations of the ability was meted the basic requirements. Content validity and structural reliability was relatively high, more comprehensive and systematic response relationship between the level of foreign students adaptability influencing factors and adaptability.3) Association rules introduced in the study of foreign students’cross-cultural adaptation. Used Visual C++development environment to achieve the Apriori algorithm, mining2-item sets frequent rules. Setting the minsup and the mine of as0.1and0.6, the according to the different degrees of support and confidence, carried out a total of17experiments, and the experimental results obtained524association rules. 4) Compared results of the frequent2-item sets Rules with traditional analysis methods, the results showed as follows:Frequent2-item sets association rules can be a good expression of the traditional analysis, and can even get interrelation which traditional analysis cannot show. Traditional analysis can be seen as part of the association analysis. Association rules can be effectively applied to the analysis of the psychological measurement data relationship.5) Using WEKA software to mining frequent a number of sets of rules, carried out6experiments. According to the different degrees of support and confidence, obtained different numbers of association rules. The study found that frequent a number of sets are better than the frequent2-item sets, it can get more information.6) Using the decision tree method for the study of foreign students’cross-cultural adaptation, selected the CART algorithm to classify the foreign students’adaptability. It established four CART trees model named a general adaptation, social and cultural adaptation, psychological adaptation and campus adaptation, and extracted twenty-nine classification rules.7) Compared the CART classification rules with traditional binary logistic regression analysis, and verify the performance of the two classification model through correct classification rate and ROC curve. The results showed that:Two classification models had their pros and cons, the test parameters were also very similar. The CART classification got the same factors with Logistic regression in mining the main effect factors and getting rules. They mutual authenticated that service mode, extraversion, image of teachers and learning conditions were the factor s with the best predictive effect. So the decision tree method was effective for the psychometric data classification, and the CART classification can get more variables than Logistic regression, it had great relationship with more sophisticated of the decision tree mining, which can avoid argument collinearity.更多还原